# ThriveOnDev Docs (full text) > ThriveOnDev documentation. Hand Linear issues to AI coding agents, get pull requests reviewed by a second agent, and follow every run. --- # Run AI coding agents on your Linear backlog > ThriveOnDev documentation. Hand Linear issues to AI coding agents, get pull requests reviewed by a second agent, and follow every run. ThriveOnDev turns a Linear issue into a reviewed GitHub pull request. An agent implements the change, a separate run reviews it, and you follow each step from Linear and your workspace. [Book a pilot call](https://cal.com/ivanzotov/pilot-intro) · [Sign in](https://thriveondev.com/orchestration/) ## Choose where to start - [Get started](https://thriveondev.com/docs/get-started/what-is-thriveondev/): What ThriveOnDev does, the terms it uses, and how one run goes from issue to merge. - [Work with the agent](https://thriveondev.com/docs/work/start-a-task/): Hand an issue to the agent, write issues it can finish, and answer it when it needs you. - [Set up your workspace](https://thriveondev.com/docs/workspace/settings/): Connect Linear and GitHub, add AI access, and choose the agents your team uses. - [Security](https://thriveondev.com/docs/security/data-and-isolation/): Where code runs, which credentials a run gets, and how ThriveOnDev can be deployed. ## Popular tasks - [Start a task from Linear](https://thriveondev.com/docs/work/start-a-task/): Assign the agent, comment, or add a label. - [Write issues an agent can finish](https://thriveondev.com/docs/work/write-good-issues/): What to put in an issue before you hand it over. - [Choose an agent with aliases](https://thriveondev.com/docs/agents/aliases/): Pick Claude Code or Codex for each task. - [Review loop and human approval](https://thriveondev.com/docs/work/review-and-approval/): How reviews repeat and when a person decides. - [Follow a run](https://thriveondev.com/docs/work/follow-a-run/): Find progress, review findings, and the result. - [Troubleshooting](https://thriveondev.com/docs/reference/troubleshooting/): What to do when a task does not start or stalls. --- Source: https://thriveondev.com/docs/ Last updated: 2026-11-20 --- # What is ThriveOnDev > ThriveOnDev runs Claude Code and Codex on your Linear issues and returns GitHub pull requests that a second agent has already reviewed. ThriveOnDev runs AI coding agents on your Linear backlog. You hand it an issue, an agent implements the change in an isolated environment, and you get a GitHub pull request back. A separate run reviews that pull request before it is merged. 1. **Implement**: An agent makes the change in an isolated environment, runs the checks and opens a pull request. 2. **Review**: A separate run, on the reviewer agent your team chose, checks the pull request against the issue. 3. **Fix review**: When the review asks for changes or release checks fail, the agent fixes the pull request and it goes back to review. 4. **Release**: The reviewed pull request must pass CI. Then it is squash-merged and the issue moves to Done. ## Who it is for Engineering teams that plan their work in Linear, keep their code in GitHub, and want coding agents to take on well-defined issues. You keep working in the tools you already use: the agent takes its tasks from Linear, reports back in the issue, and delivers to GitHub. ## What you get - **A pull request for each issue.** The agent works from the issue description, runs your checks, and opens a pull request with what it changed and how to verify it. - **A review before merge.** A second run checks every pull request against the issue. Your team picks the reviewer agent, so Codex can review Claude Code's work and the other way round. See [Review with a second agent](https://thriveondev.com/docs/agents/cross-agent-review/). - **A loop with a limit.** When the review asks for changes, the agent fixes them and the pull request is reviewed again. After three rounds the agent stops and a person decides. See [Review loop and human approval](https://thriveondev.com/docs/work/review-and-approval/). - **Your statuses, your board.** Each issue moves through your Linear statuses as the work progresses, and the [Workflow board](https://thriveondev.com/docs/workspace/workflow-board/) shows every task the agent has. - **Your AI access.** Runs use the Claude or Codex access that the people on your team add. ThriveOnDev does not resell model usage. See [Add your AI access](https://thriveondev.com/docs/agents/ai-access/). ## What stays with your team People decide what to build and write the issues. People answer the agent's questions in Linear. Owners and admins choose which repositories the agent may change, and can require a human approval before any pull request is merged. ## How you work with it 1. Connect Linear and GitHub in your [workspace settings](https://thriveondev.com/docs/workspace/settings/) and choose repositories. 2. Each person who gives tasks adds their AI access. 3. Hand an issue to the agent from Linear. See [Start a task from Linear](https://thriveondev.com/docs/work/start-a-task/). 4. Follow the run in Linear and on the Workflow board, and review the pull request in GitHub. ## Next steps - [Core concepts](https://thriveondev.com/docs/get-started/core-concepts/) - [How a run works](https://thriveondev.com/docs/get-started/how-a-run-works/) --- Source: https://thriveondev.com/docs/get-started/what-is-thriveondev/ Last updated: 2026-11-20 --- # Core concepts > The few ideas behind ThriveOnDev, including tasks and runs, stages, the review loop, aliases, AI access, slots and the workspace. Six ideas explain how ThriveOnDev behaves: a task is made of runs, runs go through stages, review and fix review form a loop, aliases choose the agent, AI access pays for the model, and slots limit how many agents run at once. ## Tasks and runs A [task](https://thriveondev.com/docs/reference/glossary/#task) is one Linear issue you handed to ThriveOnDev. A [run](https://thriveondev.com/docs/reference/glossary/#run) is one start of an agent for one stage of that task. A typical task has at least two runs: one that implements the change and one that reviews it. Every run starts in a fresh, isolated environment with access only to the repository the task targets. ## Stages Each task moves through four stages, and the Linear issue moves through your statuses with it: | Stage | What happens | Linear status | | ---------- | -------------------------------------------------------------------------- | --------------------------- | | Implement | The agent makes the change and opens a pull request. | In Progress | | Review | A separate run checks the pull request. | In Review | | Fix review | The agent addresses the findings, then the pull request is reviewed again. | In Progress, then In Review | | Release | CI must pass, then the pull request is squash-merged. | In Release, then Done | If your workspace requires [human approval](https://thriveondev.com/docs/reference/glossary/#human-approval), the task waits in **In Human Review** between review and release. ## The review loop Review and fix review repeat until the review approves. The loop has a limit: after three automatic fix-review rounds the agent stops and leaves a comment in Linear, and a person decides what to do next. Read more in [Review loop and human approval](https://thriveondev.com/docs/work/review-and-approval/). ## Aliases An [alias](https://thriveondev.com/docs/reference/glossary/#alias) such as `#opus` or `#sol` names an agent and a model. Write one in an issue to choose who implements it. Your workspace also has a default developer agent and a reviewer agent. See [Choose an agent with aliases](https://thriveondev.com/docs/agents/aliases/). ## AI access Runs use the [AI access](https://thriveondev.com/docs/reference/glossary/#ai-access) of the person who gave the task: their Claude key or their ChatGPT sign-in for Codex. Tasks picked up through the `Automatable` label use the access of the person your workspace chose for automatic tasks. ## Slots and the queue Your workspace runs a limited number of agents at the same time. Each running agent takes a [slot](https://thriveondev.com/docs/reference/glossary/#slot). When all slots are busy, new tasks wait in the [queue](https://thriveondev.com/docs/reference/glossary/#queue) and start by themselves. ## Workspace Your [workspace](https://thriveondev.com/docs/reference/glossary/#workspace) holds your connections to Linear and GitHub, the repositories the agent may change, your people and their roles, and your settings. It has two screens: **Workflow** and **Settings**. ## Next steps - [How a run works](https://thriveondev.com/docs/get-started/how-a-run-works/) - [Glossary](https://thriveondev.com/docs/reference/glossary/) --- Source: https://thriveondev.com/docs/get-started/core-concepts/ Last updated: 2026-11-20 --- # How a run works > Follow one Linear issue through ThriveOnDev, from implement to review, fix review and release, with the Linear status at each step. A ThriveOnDev task goes from a Linear issue to a merged pull request in four stages: implement, review, fix review when needed, and release. Each stage is a separate [run](https://thriveondev.com/docs/reference/glossary/#run), and the Linear issue shows where the task is. 1. **Implement**: An agent makes the change in an isolated environment, runs the checks and opens a pull request. 2. **Review**: A separate run, on the reviewer agent your team chose, checks the pull request against the issue. 3. **Fix review**: When the review asks for changes or release checks fail, the agent fixes the pull request and it goes back to review. 4. **Release**: The reviewed pull request must pass CI. Then it is squash-merged and the issue moves to Done. ## 1. Implement When you hand an issue to the agent, the issue moves to **In Progress**. The agent reads the issue and the linked context, clones the target repository in an isolated environment, makes the change and runs the relevant checks. It then opens a pull request that says what changed and how to verify it. If something in the issue is unclear, the agent asks in the Linear thread and waits for your answer. The task shows as **Needs you** on the Workflow board until you reply. ## 2. Review The issue moves to **In Review**, and a separate run checks the pull request against the issue. That run uses your workspace's [reviewer agent](https://thriveondev.com/docs/agents/cross-agent-review/), which can be a different agent from the one that wrote the code. The review either approves the pull request or requests changes. ## 3. Fix review When the review requests changes, the issue returns to **In Progress** and the agent addresses the findings on the same pull request. The pull request then goes back to review. This loop has a limit. After three automatic fix-review rounds, the next rejection pauses the loop: the agent leaves a comment, the issue stays in **In Review**, and a person decides what to do. Change requests from people don't count toward the limit. ## Optional: human approval If your workspace requires human approval, an approved pull request moves to **In Human Review** and waits for an approving review in GitHub. If a person requests changes, the task goes back to fix review. See [Review loop and human approval](https://thriveondev.com/docs/work/review-and-approval/). ## 4. Release The issue moves to **In Release**. ThriveOnDev checks that the pull request's CI passes, squash-merges the exact commit that was reviewed, deletes the branch, and moves the issue to **Done**. Release never rewrites the branch. If CI fails or the pull request conflicts with the base branch, the task goes back to fix review, where the agent merges the base branch in and fixes the problem. A task can return from release this way up to three times. ## Limits that apply to every run - A run stops after 5 hours. Split larger work into several issues. - A run can only change the repositories and branches your workspace allows. - If all of your workspace's slots are busy, the task waits in the [queue](https://thriveondev.com/docs/work/queue-and-capacity/) and starts by itself. ## Next steps - [Run your first task](https://thriveondev.com/docs/get-started/first-task/) - [Review loop and human approval](https://thriveondev.com/docs/work/review-and-approval/) --- Source: https://thriveondev.com/docs/get-started/how-a-run-works/ Last updated: 2026-11-20 --- # Run your first task > Check your workspace setup, pick a small Linear issue, hand it to the agent, and review the pull request it opens. Your first task should be small and easy to check: one change that fits in one pull request. This page assumes your workspace already exists. You can open it at `thriveondev.com/orchestration`. 1. **Check your setup** Open **Settings** in your workspace. All four stations should be connected: **Linear**, **AI models**, **GitHub** and **Team**. At least one repository must be chosen under GitHub, and you need your own AI access and the **I can give tasks** permission. Settings marks the next missing step. See [Workspace settings](https://thriveondev.com/docs/workspace/settings/). 2. **Pick a small issue** Choose a change you could review in a few minutes, such as a clearer error message or a small fix with an obvious test. Write what should change, where, and how you will check it. See [Write issues an agent can finish](https://thriveondev.com/docs/work/write-good-issues/). 3. **Hand it to the agent** In Linear, assign or delegate the issue to the ThriveOnDev agent. The agent replies "Starting to work on this issue…" and the issue moves to **In Progress**. To choose a specific agent, add an alias such as `#opus` or `#sol` to the issue first. 4. **Answer if asked** If the agent needs a detail, it asks in the Linear thread. Reply there. The task shows as **Needs you** on the Workflow board until you do. 5. **Review the pull request** When the issue reaches **In Review**, the agent has opened a pull request and a review run is checking it. Open the pull request from the issue or from the task card on the Workflow board, and compare the change with what you asked for. > **Note:** If the agent replies that it is off for this workspace, ThriveOnDev has not turned it on yet. Contact us at . ## What to expect A small task usually ends with a reviewed pull request that is merged once CI passes. If your workspace requires human approval, it waits in **In Human Review** for your approval in GitHub first. If something goes wrong, the agent explains it in Linear. See [Troubleshooting](https://thriveondev.com/docs/reference/troubleshooting/). ## Next steps - [Write issues an agent can finish](https://thriveondev.com/docs/work/write-good-issues/) - [Follow a run](https://thriveondev.com/docs/work/follow-a-run/) --- Source: https://thriveondev.com/docs/get-started/first-task/ Last updated: 2026-11-20 --- # Start a task from Linear > Hand a Linear issue to ThriveOnDev by assigning the agent, mentioning it, writing a stage command, or adding the Automatable label. To start a task, assign or delegate a Linear issue to the ThriveOnDev agent. You can also mention the agent in a comment, write a stage command such as `/implement`, or let ThriveOnDev pick up issues labelled `Automatable`. ## Before you start - Your workspace has Linear and GitHub connected, and at least one repository is chosen. See [Workspace settings](https://thriveondev.com/docs/workspace/settings/). - You have added your own [AI access](https://thriveondev.com/docs/agents/ai-access/). The run uses it. - An owner or admin has given you **I can give tasks** under **Settings → Team**. ## Assign or delegate the issue Open the issue in Linear and assign or delegate it to the ThriveOnDev agent. The agent replies "Starting to work on this issue…", the issue moves to **In Progress**, and the implement stage begins. This is the most common way to start. The agent works on the issue's description, comments and linked issues. ## Mention the agent If the issue is already assigned to a person, mention the agent in a comment instead. It starts the same way as an assignment. ## Write a stage command A comment line that starts with `/implement`, `/review`, `/fix-review` or `/release` runs that stage. Use it to run a stage again, for example a fresh review after you changed the pull request yourself: ```text @thriveondev /review ``` See [Linear commands and labels](https://thriveondev.com/docs/reference/linear-commands/) for every command. ## Choose the agent for this task Add an [alias](https://thriveondev.com/docs/reference/glossary/#alias) such as `#opus` or `#sol` to the issue or to your comment. Without one, your workspace's developer agent does the work. See [Choose an agent with aliases](https://thriveondev.com/docs/agents/aliases/). ## Let ThriveOnDev pick up issues automatically Add the `Automatable` label to an issue that is not started yet. When **Automatic tasks** is on under **Settings → Linear**, ThriveOnDev checks for labelled issues about once a minute and takes them on. It skips issues whose blocking issues are not finished yet, and comes back to them later. Automatic tasks run on the AI access of one person your workspace chose, and that person must allow it with **Agents may use my AI access**. Owners and admins can also limit which Linear teams are checked. ## What happens next The agent works through the stages in [How a run works](https://thriveondev.com/docs/get-started/how-a-run-works/). If all of your workspace's slots are busy, the issue gets the `queued` label and a comment with its place in line, and it starts by itself. See [Queue and capacity](https://thriveondev.com/docs/work/queue-and-capacity/). To stop a run, press the stop button on the agent session in Linear. ## Next steps - [Write issues an agent can finish](https://thriveondev.com/docs/work/write-good-issues/) - [Linear commands and labels](https://thriveondev.com/docs/reference/linear-commands/) --- Source: https://thriveondev.com/docs/work/start-a-task/ Last updated: 2026-11-20 --- # Write issues an agent can finish > What to put in a Linear issue before you hand it to an AI coding agent, with a template, a good example, and the size that works best. An agent finishes an issue when it knows what to change, where, and how to tell it is done. Write the issue the way you would for a new teammate who can read the code but was not in your planning meeting. ## Include these four things 1. **The outcome.** What should be true when the work is done, in one or two sentences. 2. **Where.** The repository, and the screen, endpoint, module or file area if you know it. 3. **How to check it.** The acceptance criteria: what a reviewer should see, and which tests or commands prove it. 4. **Context.** Links to related issues, designs, error logs or earlier pull requests. The agent reads linked Linear issues. ## Template ```markdown ## Goal Show a clear message when an export has no rows, instead of an empty file. ## Where Repo: acme/web-app Base branch: main Export button on the Reports page (src/features/reports). ## Acceptance criteria - An empty export shows "Nothing to export for these filters". - A non-empty export still downloads a CSV. - Unit tests cover both cases. ## Context Reported in ACME-412. ``` The `Repo:` and `Base branch:` lines are only needed when your workspace allows more than one repository, or when the target is not obvious. Without them, the agent asks. ## Keep it to one pull request Good first tasks are changes a reviewer can check in a few minutes. A run stops after 5 hours, so split large work into several issues and connect them as blocking issues in Linear. Issues picked up through the `Automatable` label wait until the issues blocking them are done. ## Make review possible The review run checks the pull request against what the issue asks for, so vague criteria make for vague reviews. Name the behavior to verify, not just the code to touch. "Fix the export" gives the reviewer nothing to check. "Empty exports show a message and tests cover it" does. ## Mind the hashtags Any `#word` that matches an alias, such as `#sol`, chooses that agent. Use aliases on purpose. See [Choose an agent with aliases](https://thriveondev.com/docs/agents/aliases/). ## When the agent asks If something is missing, the agent asks in the Linear thread and waits. Answer there. The clearer the issue, the fewer questions you get. ## Next steps - [Start a task from Linear](https://thriveondev.com/docs/work/start-a-task/) - [Follow a run](https://thriveondev.com/docs/work/follow-a-run/) --- Source: https://thriveondev.com/docs/work/write-good-issues/ Last updated: 2026-11-20 --- # Follow a run > See where a ThriveOnDev task is, in the Linear issue, on the Workflow board, and in the GitHub pull request, and know when it needs you. You can follow a task in three places. The Linear issue shows its status and the agent's messages. The [Workflow board](https://thriveondev.com/docs/workspace/workflow-board/) shows every task at a glance. The GitHub pull request shows the code and the checks. ## In Linear - **The status** shows the stage: **In Progress** while the agent implements or fixes, **In Review** while a review runs, **In Human Review** while it waits for your approval, **In Release** while CI is checked and the pull request is merged, and **Done**. - **The thread** shows what the agent is doing, its questions, and any problem it hit. - **The `queued` label** and a "⏳ **Queued**" comment mean the task is waiting to start. The comment shows the reason and its position. ## On the Workflow board Each task is a ball in the column of its Linear status. Its look tells you what the agent is doing: | State | Meaning | | ----------------------------- | ----------------------------------------------------------------- | | **Waiting for an agent** | In the queue, or about to start. | | **An agent is working on it** | A run is in progress. The card says which agent and for how long. | | **Needs you** | Waiting for your answer, your review or your AI access. | | **Nothing is happening** | Open in Linear, but no agent is working on it. | | **Didn't work out** | The run stopped. The card says why and what is needed. | | **Done** | Merged. | Click a task to open its card. The card shows what changed, how to check it, what is needed from you, and the pull requests, with buttons to **Review in GitHub**, **Answer in Linear** or **Open in Linear**. ## In GitHub The pull request has the diff, the description of what changed and how to verify it, the CI checks, and the review findings. If your workspace requires human approval, your approving review there is what lets the task continue. ## When a task needs you | You see | Do this | | ------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------ | | A question in the Linear thread | Reply in the thread. | | **In Human Review** | Review the pull request in GitHub. Approve it, or request changes. | | "**Automatic fix-review paused.**" | The review loop hit its limit. Clarify the issue or fix the pull request, then comment `/fix-review` or `/review`. | | A stop reason on the card, such as "AI access not connected" | Follow **What is needed** and **Open Settings** on the card. | See [Troubleshooting](https://thriveondev.com/docs/reference/troubleshooting/) for every stop reason. ## Next steps - [The Workflow board](https://thriveondev.com/docs/workspace/workflow-board/) - [Review loop and human approval](https://thriveondev.com/docs/work/review-and-approval/) --- Source: https://thriveondev.com/docs/work/follow-a-run/ Last updated: 2026-11-20 --- # Review loop and human approval > How ThriveOnDev reviews every pull request, when the review loop stops and hands the decision to a person, and how to require human approval before merge. Every pull request ThriveOnDev opens is reviewed by a separate run before it can be merged. If the review asks for changes, the agent fixes them and the pull request is reviewed again, up to three automatic rounds. Owners and admins can also require a person's approval before any merge. ## How the review loop works 1. The implement run opens a pull request, and the issue moves to **In Review**. 2. A review run checks the pull request against the issue, using your workspace's [reviewer agent](https://thriveondev.com/docs/agents/cross-agent-review/). 3. If the review approves, the task moves on to release, or to human approval if your workspace requires it. 4. If the review requests changes, the issue moves back to **In Progress**. A fix-review run addresses the findings on the same pull request, and then it is reviewed again. ## When the loop stops After three automatic fix-review rounds, the next rejection pauses the loop. The agent leaves a comment that starts with "**Automatic fix-review paused.**" and starts no new session. The issue stays in **In Review**. Nothing else changes: the assignee, labels and status stay as they are. A person decides what happens next. Usually you do one of these: - **Clarify the issue** if the review and the implementation disagree about what was asked, then comment `/fix-review`. - **Fix the pull request yourself**, then comment `/review` for a fresh review. - **Close the pull request** if the task is not worth continuing. Change requests from people don't count toward the three rounds. The count starts again when the task reaches human approval, release or done. ## Require human approval before release By default, a pull request that passes review is merged once its CI checks pass. To have a person approve every pull request first: 1. Open **Settings → Team**. 2. Turn on **Require human approval before release**. Only owners and admins can change this setting. When it is on: - An approved pull request moves to **In Human Review**. ThriveOnDev creates this Linear status if your team does not have it. - The task waits for an approving review in GitHub. - If the person requests changes in GitHub, the task goes back to fix review. Turning it off asks you to confirm **Release without human approval?**, because pull requests will then merge without a person. ## What review checks The review run reads the issue and the pull request and checks that the change does what the issue asks and that it is safe to merge. The clearer your acceptance criteria, the more useful the review. See [Write issues an agent can finish](https://thriveondev.com/docs/work/write-good-issues/). ## Next steps - [Review with a second agent](https://thriveondev.com/docs/agents/cross-agent-review/) - [Follow a run](https://thriveondev.com/docs/work/follow-a-run/) --- Source: https://thriveondev.com/docs/work/review-and-approval/ Last updated: 2026-11-20 --- # Queue and capacity > Why a ThriveOnDev task waits in the queue, how slots limit the agents running at once, and how waiting tasks start again on their own. Your workspace runs a limited number of agents at the same time. Each running agent takes one [slot](https://thriveondev.com/docs/reference/glossary/#slot). When every slot is busy, or a task is waiting for something else, the task joins the queue and starts by itself as soon as it can. ## Why a task waits | Reason | Starts when | | -------------------------------------- | ----------------------------------------------------------------------------------------------------------- | | All of your workspace's slots are busy | Another of your tasks finishes or is cancelled. | | The platform is at capacity | Room frees up on the platform. | | Its blocking issues are not done | The issues that block it in Linear are finished. | | It is waiting for CI checks | The pull request's checks finish. A task blocked by failing checks waits up to 7 days for a re-run to pass. | | The AI provider is unavailable | The provider responds again. | | It is on hold | An admin releases it. | ## How you see it - **In Linear**, the issue gets the `queued` label and a comment such as "⏳ **Queued** — … Position 2. Resumes automatically when …". When it starts, the agent comments "▶️ Resumed after …". - **On the Workflow board**, the **To do** list shows how many tasks are waiting for an agent and their order: **Next**, **2nd**, **3rd** and so on. If you send the agent a message while the task is still queued, it replies that the message was not delivered yet. Send it again once the task has started. ## How waiting tasks start ThriveOnDev checks the queue about once a minute and starts the next task that can run. If a task fails to start three times, the agent comments "⚠️ Could not start …" and explains why. ## Cancel a waiting task Any member can cancel a task in the **To do** list: choose **Cancel task** and confirm. To stop a task that is already running, press stop on the agent session in Linear. ## How many slots you have The number of slots is set per workspace. Shared slots are why one long task can delay others: splitting work into smaller issues keeps the queue moving. If your team regularly waits for slots, contact ThriveOnDev at . ## Next steps - [Follow a run](https://thriveondev.com/docs/work/follow-a-run/) - [Troubleshooting](https://thriveondev.com/docs/reference/troubleshooting/) --- Source: https://thriveondev.com/docs/work/queue-and-capacity/ Last updated: 2026-11-20 --- # Supported coding agents > ThriveOnDev runs Claude Code and Codex. See which models each alias uses and how to choose the developer and reviewer agents for your team. ThriveOnDev runs two coding agents: **Claude Code** from Anthropic and **Codex** from OpenAI. Each person on your team connects the one they use, and your workspace decides which agent implements and which reviews. ## Agents and their aliases | Agent | Alias | Model | | ----------- | -------- | ---------------- | | Claude Code | `#opus` | Claude Opus 5.5 | | Claude Code | `#fable` | Claude Fable 5.1 | | Codex | `#astra` | GPT-6.2 Astra | | Codex | `#sol` | GPT-6.2 Sol | `#astra` is the platform default. Your workspace can choose a different default. See [Choose an agent with aliases](https://thriveondev.com/docs/agents/aliases/). ## What you need to use an agent - **Claude Code** needs a Claude key, added under **Settings → AI models**. - **Codex** needs your ChatGPT sign-in: sign in to Codex with a paid ChatGPT account and paste its `auth.json` file. Each person adds their own. See [Add your AI access](https://thriveondev.com/docs/agents/ai-access/). ## Developer and reviewer Two workspace settings decide which agent does what when an issue does not name one: - **Developer** implements the change and fixes review findings. - **Reviewer** reviews every pull request. By default it is **Same as developer**, which means the same agent reviews in a separate run. Choose a different reviewer to have one agent check another's work. See [Review with a second agent](https://thriveondev.com/docs/agents/cross-agent-review/). ## Mixing agents You can use both agents in the same workspace. One issue can use `#opus` and the next `#sol`, while every pull request is reviewed by your chosen reviewer. Each run uses the AI access of the person who gave the task for the agent that runs it. ## Next steps - [Choose an agent with aliases](https://thriveondev.com/docs/agents/aliases/) - [Add your AI access](https://thriveondev.com/docs/agents/ai-access/) --- Source: https://thriveondev.com/docs/agents/supported-agents/ Last updated: 2026-11-20 --- # Choose an agent with aliases > Write an alias such as #opus or #sol in a Linear issue to choose the coding agent and model for that task, or set your workspace default. An alias is a short name for a coding agent and a model, such as `#opus` or `#sol`. Write it in a Linear issue or comment to choose who works on that task. Without one, your workspace's developer agent does the work. ## Built-in aliases **Claude Code** | Alias | Model | | -------- | ---------------- | | `#opus` | Claude Opus 5.5 | | `#fable` | Claude Fable 5.1 | **Codex** | Alias | Model | | -------- | ----------------------------------- | | `#astra` | GPT-6.2 Astra, the platform default | | `#sol` | GPT-6.2 Sol | ## Use an alias on one task Put the alias anywhere in the issue description or in the comment that starts the work: ```text @thriveondev #opus /implement ``` The alias applies to the implement and fix-review runs of that task. The review uses your workspace's reviewer agent. ## How the agent is chosen ThriveOnDev looks for an agent in this order: 1. The first `#word` in the issue or comment that matches an alias. 2. Your workspace's **Developer** setting. 3. The platform default, `#astra`. > **Warning:** Any `#word` that matches an alias selects that agent, even in a sentence like "fixes the #sol layout". Write aliases only when you mean them. ## Set your workspace default Owners and admins choose the default under **Settings → AI models → Developer**. It applies to every task that does not name an alias, including tasks picked up through the `Automatable` label. ## Make sure the alias can run An alias only works if the person who gave the task has AI access for that agent. A `#sol` task needs a Codex sign-in, and an `#opus` task needs a Claude key. If it is missing, the task stops with **AI access not connected**. See [Add your AI access](https://thriveondev.com/docs/agents/ai-access/). ## Next steps - [Review with a second agent](https://thriveondev.com/docs/agents/cross-agent-review/) - [Linear commands and labels](https://thriveondev.com/docs/reference/linear-commands/) --- Source: https://thriveondev.com/docs/agents/aliases/ Last updated: 2026-11-20 --- # Review with a second agent > Choose a reviewer agent so Codex reviews Claude Code's pull requests, or the other way round, and every change gets a second opinion before merge. Every pull request ThriveOnDev opens is reviewed in a separate run before it can be merged. Your workspace's **Reviewer** setting decides which agent does that review. Pick a different agent from your developer to get a second model's judgement on every change. ## Why use a different reviewer An agent reviewing its own work tends to agree with itself. A different model notices other problems, like missed edge cases, unclear names or tests that don't test the change. Because the reviewer is a workspace setting, every pull request gets it, whichever alias implemented the task. ## Set the reviewer 1. Open **Settings → AI models**. 2. Under **Reviewer**, choose an alias, for example `#sol` if your team mostly implements with `#opus`. Only owners and admins can change it. The default is **Same as developer**: the same agent reviews, but in a separate run with fresh context. ## Make sure reviews can run The review runs on the AI access of the person who gave the task. If your reviewer is Codex and that person has only connected Claude, the review stops with **AI access not connected**. Ask the people who give tasks to add access for both agents. See [Add your AI access](https://thriveondev.com/docs/agents/ai-access/). ## What the reviewer does The review run reads the issue and the pull request, checks that the change does what was asked, and either approves or requests changes. Requested changes go to a fix-review run by the developer agent, and the pull request comes back to the same reviewer. See [Review loop and human approval](https://thriveondev.com/docs/work/review-and-approval/). ## Next steps - [Review loop and human approval](https://thriveondev.com/docs/work/review-and-approval/) - [Supported coding agents](https://thriveondev.com/docs/agents/supported-agents/) --- Source: https://thriveondev.com/docs/agents/cross-agent-review/ Last updated: 2026-11-20 --- # Add your AI access > Connect your own Claude key or ChatGPT sign-in for Codex in ThriveOnDev, see which access a run uses, and renew it when it expires. Agents run on the AI access of the people who give them tasks. Each person adds their own under **Settings → AI models**: a Claude key for Claude Code, a ChatGPT sign-in for Codex, or both. ThriveOnDev does not resell model usage. Your usage is billed by Anthropic or OpenAI under your own account. > **Note:** If **Settings → AI models** says "AI access is provided by ThriveOnDev", your workspace's model usage is included and nobody needs to add access. ## Add access 1. Open **Settings → AI models** and choose **Add AI access**. 2. Pick the provider: - **Claude**: paste your key into **API key**. - **Codex**: sign in to Codex with your paid ChatGPT account, then paste the complete local `auth.json` file. 3. Save. The key is stored encrypted and is never shown again. If you have access for both, you can mark one as your default. ## Which access a run uses | How the task started | Whose access | | -------------------------------------------------------- | ----------------------------------------------------------------------------------------------------- | | You assigned, delegated, mentioned or commented | Yours | | ThriveOnDev picked it up through the `Automatable` label | The person your workspace chose for automatic tasks, who must turn on **Agents may use my AI access** | The agent that runs must match the access: an `#opus` task needs a Claude key, and a `#sol` task needs a Codex sign-in. The review uses the reviewer agent, so you may need access for both. ## Permission to give tasks Having AI access is not enough on its own. An owner or admin must also turn on **I can give tasks** for you under **Settings → Team**. See [Team and roles](https://thriveondev.com/docs/workspace/team-and-roles/). ## When access stops working The task card and the Linear thread tell you what happened: | You see | Do this | | ------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------- | | **AI access not connected** | Add access for the agent the task uses. | | **AI access rejected** or "access needs to be renewed" | Remove the old access and add it again. A new Codex `auth.json` is often needed after you sign in again. | | **AI credits used up** or **AI usage limit reached** | Your provider account has run out of usage. Top up or wait for the limit to reset, then start the task again. | ## Remove access Remove your access under **Settings → AI models** at any time. Tasks that would use it stop with **AI access not connected** until someone adds access again. ## Next steps - [Supported coding agents](https://thriveondev.com/docs/agents/supported-agents/) - [Team and roles](https://thriveondev.com/docs/workspace/team-and-roles/) --- Source: https://thriveondev.com/docs/agents/ai-access/ Last updated: 2026-11-20 --- # The Workflow board > Read the Workflow board in your ThriveOnDev workspace, where every task appears in its Linear status and its look shows what the agent is doing. The **Workflow** screen shows every task the agent has in your workspace, on one board. The columns are your Linear statuses, in your Linear colours, and each task stands where Linear says it is. How a task is drawn tells you what the agent is doing with it right now. ## Columns Tasks wait in **To do**, move through the statuses in progress, and collect in **Done**. Each column header shows a count and the status name. Hover over a column to see what happens there, for example "A second agent checks the code" for the review status. The board refreshes on its own. The badge in the header reads **Live**, or **Agent off** if the agent has not been turned on for your workspace. ## Task states **Waiting for an agent** The task is in the queue or about to start. **An agent is working on it** A run is in progress. The card shows which agent and for how long, for example "Claude is working · 25 min". **Needs you** The task waits for you: an answer in Linear, your review in GitHub, or your AI access. The board highlights one such task at a time with a button that takes you where to act. **Nothing is happening** The issue is in a working status in Linear, but no agent is on it. **Didn't work out** The run stopped. The card names the reason and what is needed. **Done** The pull request was merged. ## Task card Click a task to open its card. It shows: - the current state, for example "Waiting for your review · 2 h"; - which agent made the change; - **What changed** and **How to check**, taken from the pull request; - **What is needed**, when the task waits for something; - **Pull requests**, each with a state such as **Waiting for your review** or **Merged**. Buttons take you where to act: **Review in GitHub**, **Answer in Linear** or **Open in Linear**. When a task stopped because of a setup problem, **Open Settings** takes you to the station that needs attention. The board deliberately shows no costs and no technical logs. It answers one question: where is each task, and does anything need you. ## The To do list Open **To do** to see the tasks waiting for an agent, in order: **Next**, **2nd**, **3rd** and so on. Any member can cancel a waiting task with **Cancel task**. See [Queue and capacity](https://thriveondev.com/docs/work/queue-and-capacity/). ## Next steps - [Follow a run](https://thriveondev.com/docs/work/follow-a-run/) - [Workspace settings](https://thriveondev.com/docs/workspace/settings/) --- Source: https://thriveondev.com/docs/workspace/workflow-board/ Last updated: 2026-11-20 --- # Workspace settings > Set up your ThriveOnDev workspace in Settings, where four stations connect Linear, AI models, GitHub and your team, and show what is still missing. **Settings** is where your workspace is connected to the tools a task passes through. It is one track with four stations, in the order a task travels: **Linear**, **AI models**, **GitHub** and **Team**. Each station shows its status and one line of detail. Click a station to open its card. ## The four stations | Station | What it sets up | Who can change it | | -------------------------------- | --------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------- | | **Linear**: Tasks come from | The Linear workspace, which teams the agent works in, and **Automatic tasks** for the `Automatable` label | Owners and admins | | **AI models**: Agents think with | The **Developer** and **Reviewer** agents, and each person's own AI access | Agents: owners and admins. AI access: each person | | **GitHub**: Code goes to | The GitHub App installation and the repositories the agent may change | Owners and admins | | **Team**: People who give tasks | Who may give tasks, each person's Linear link and consent, and **Require human approval before release** | Workspace settings: owners and admins. Your own link and consent: you | A station with a problem gets a highlighted ring and one callout that explains the fix. ## Set up a workspace Until the setup is finished, the workspace opens on Settings and marks the next step. Owners and admins usually do it in this order: 1. **Connect Linear** Authorize the ThriveOnDev app in your Linear workspace, then optionally choose which teams it works in. See [Linear integration](https://thriveondev.com/docs/integrations/linear/). 2. **Connect GitHub** Install the ThriveOnDev GitHub App on your GitHub organization. See [GitHub integration](https://thriveondev.com/docs/integrations/github/). 3. **Choose repositories** Pick the repositories, and the branches, the agent may change. Nothing is allowed until you choose. 4. **Link your Linear account** Each person links their own Linear user, so ThriveOnDev knows who gave a task. 5. **Add AI access** Each person who gives tasks adds a Claude key or a Codex sign-in. See [Add your AI access](https://thriveondev.com/docs/agents/ai-access/). 6. **Give an agent its first task** Hand a small issue to the agent in Linear. See [Run your first task](https://thriveondev.com/docs/get-started/first-task/). Once Linear, GitHub and at least one repository are set up, the workspace opens on the **Workflow** board instead. A dot next to **Settings** in the sidebar stays until everything is done. ## Change settings later Open the station and change what you need. Members see every station, but controls they cannot use are locked and say "Only owners and admins can change this." See [Team and roles](https://thriveondev.com/docs/workspace/team-and-roles/). ## Next steps - [Linear integration](https://thriveondev.com/docs/integrations/linear/) - [GitHub integration](https://thriveondev.com/docs/integrations/github/) --- Source: https://thriveondev.com/docs/workspace/settings/ Last updated: 2026-11-20 --- # Team and roles > What owners, admins and members can do in a ThriveOnDev workspace, and how to let someone give tasks to the agent. A ThriveOnDev workspace has three roles: **owner**, **admin** and **member**. Everyone sees the same screens. Roles only change what a person can change. Owners and admins have the same rights. ## What each role can do | Action | Owner or admin | Member | | -------------------------------------------------------- | -------------- | ------ | | See the Workflow board and Settings | Yes | Yes | | Link their own Linear account | Yes | Yes | | Add or remove their own AI access | Yes | Yes | | Allow agents to use their AI access for automatic tasks | Yes | Yes | | Cancel a waiting task | Yes | Yes | | Connect Linear and GitHub, choose teams and repositories | Yes | No | | Choose the developer and reviewer agents | Yes | No | | Turn **Automatic tasks** on or off | Yes | No | | Give people permission to give tasks | Yes | No | | Require human approval before release | Yes | No | Locked controls tell members "Only owners and admins can change this." ## Let someone give tasks A person can hand issues to the agent only when an owner or admin has turned on **Can give tasks** for them under **Settings → Team**. They also need their own [AI access](https://thriveondev.com/docs/agents/ai-access/), because their tasks run on it. ## Choose who runs automatic tasks Tasks picked up through the `Automatable` label have no person behind them, so they run on the AI access of one chosen person. That person must turn on **Agents may use my AI access** under **Settings → Team**. The Linear station shows whose access automatic tasks use. ## Link your Linear account Each person links their Linear user under **Settings → Team → Your Linear**. This read-only link tells ThriveOnDev who gave a task, so the run uses the right AI access. Unlink it at any time. ## Next steps - [Add your AI access](https://thriveondev.com/docs/agents/ai-access/) - [Workspace settings](https://thriveondev.com/docs/workspace/settings/) --- Source: https://thriveondev.com/docs/workspace/team-and-roles/ Last updated: 2026-11-20 --- # Linear integration > Connect ThriveOnDev to Linear, choose the teams the agent works in, turn on automatic tasks, and see which permissions the app uses. ThriveOnDev takes its tasks from Linear. An owner or admin connects your Linear workspace once. After that the ThriveOnDev agent can be assigned, delegated and mentioned in your issues, and it reports progress in the issue thread. ## Connect Linear 1. Open **Settings** and select the **Linear** station. 2. Choose **Connect Linear** and approve the ThriveOnDev app in Linear. 3. Optionally, choose **Choose teams** to limit which Linear teams the agent works in. With no teams chosen, it works in all of them. The app is installed as an agent in Linear, so it appears in the assignee list and can be mentioned. ## Permissions | Permission | Why | | ----------- | ------------------------------------------------------------ | | Read | Read issues, comments and linked issues to understand a task | | Write | Comment on issues and move them through your statuses | | Assignable | Let you assign and delegate issues to the agent | | Mentionable | Let you mention the agent in comments | Each person also links their own Linear user under **Settings → Team** with a separate, read-only sign-in. This tells ThriveOnDev who gave a task. ## Statuses the agent uses As a task moves through its stages, the agent moves the issue through these statuses: **In Progress**, **In Review**, **In Human Review** (only when human approval is on), **In Release** and **Done**. If your team does not have **In Human Review**, ThriveOnDev creates it when it is first needed. Child issues move with their parent. ## Automatic tasks **Automatic tasks** lets ThriveOnDev pick up unstarted issues labelled `Automatable` without anyone assigning them. Turn it on in the Linear station. Its state is **On**, **Off** or **Paused**. It checks the chosen teams about once a minute and skips issues whose blocking issues are not done yet. See [Start a task from Linear](https://thriveondev.com/docs/work/start-a-task/). ## Change or disconnect Owners and admins can connect a different Linear workspace from the Linear station. To remove ThriveOnDev's access completely, uninstall the app in your Linear workspace settings. ## Next steps - [Start a task from Linear](https://thriveondev.com/docs/work/start-a-task/) - [GitHub integration](https://thriveondev.com/docs/integrations/github/) --- Source: https://thriveondev.com/docs/integrations/linear/ Last updated: 2026-11-20 --- # GitHub integration > Install the ThriveOnDev GitHub App, choose exactly which repositories and branches agents may change, and see how pull requests are merged. ThriveOnDev delivers its work to GitHub as pull requests. An owner or admin installs the ThriveOnDev GitHub App and then chooses the repositories the agent may change. Installing the app alone allows nothing. ## Connect GitHub 1. Open **Settings** and select the **GitHub** station. 2. Choose **Connect GitHub** and install the ThriveOnDev GitHub App on your organization. You can give it all repositories or selected ones. 3. Choose **Choose repositories** and pick the repositories the agent may work in. ## Choose repositories and branches The agent works only in repositories on your allow-list. For each repository you can set which base branches it may target. The default is `main`. A task that targets anything else stops with **Repository not allowed**. If you allow several repositories, tell the agent which one a task is for by adding these lines to the issue: ```text Repo: your-org/your-repo Base branch: main ``` If a repository later becomes unreadable for the app, for example because it was removed from the installation, it shows as **Unavailable** with a **Remove** button. ## What the agent does in GitHub - Creates a branch and opens a pull request for each task. - Pushes fixes to the same pull request after a review asks for changes. - At release, checks that CI passes and squash-merges the exact commit that was reviewed, then deletes the branch. Release never rebases or rewrites the branch. When the base branch has moved on, a fix-review run merges it into the pull request instead. If your workspace requires human approval, the pull request waits for an approving review from a person on your team before release. See [Review loop and human approval](https://thriveondev.com/docs/work/review-and-approval/). ## Credentials Each run gets a short-lived GitHub token that is valid only for the repositories your workspace allows. Tokens last one hour and are renewed during a long run only while the repository is still allowed. See [Data handling and isolation](https://thriveondev.com/docs/security/data-and-isolation/). ## Next steps - [Data handling and isolation](https://thriveondev.com/docs/security/data-and-isolation/) - [How a run works](https://thriveondev.com/docs/get-started/how-a-run-works/) --- Source: https://thriveondev.com/docs/integrations/github/ Last updated: 2026-11-20 --- # Data handling and isolation > Where ThriveOnDev runs your code, which short-lived credentials each run gets, what data is kept and for how long, and its assurance status. Each ThriveOnDev run happens in its own short-lived job, in an environment reserved for your workspace. The job gets only the credentials that one run needs, for only the repositories you allowed, and it is deleted shortly after it finishes. ## Where a run happens - **One job per run.** Every stage of every task, whether implement, review, fix review or release, starts as a separate job with a fresh copy of the repository. - **A space per workspace.** Your workspace's jobs run in their own Kubernetes namespace with resource limits. Jobs from other workspaces cannot reach them. - **Network rules.** A job accepts connections only from the ThriveOnDev control plane. It can reach the internet, for example to install dependencies or call the AI provider, but not private or internal network ranges. - **Time limit.** A run is stopped after 5 hours. - **Clean-up.** The repository copy exists only inside the job. Finished jobs are deleted after about 30 seconds. ## Credentials a run gets | Credential | Scope and lifetime | | ---------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | GitHub | An app installation token limited to the repositories your workspace allows. It lasts one hour and is renewed during a long run only while the repository is still allowed. | | AI access | The Claude key or Codex sign-in of the person who gave the task. Keys are stored encrypted (AES-GCM) and are never shown again after you save them. | | Delivery | A one-time token that hands the job its credentials and expires within 15 minutes. | Callbacks from a job to ThriveOnDev are signed with HMAC-SHA256. ## What is kept | Data | Kept for | | ------------------------------------------------------ | ------------------------------------------------------------------------ | | Repository contents | The life of the job only | | Linear issue and comment details the agent used | The service term | | Run history: stages, outcomes and failure reasons | Workflow history 7 days, events 3 months, other records the service term | | Agent session records: prompts, output and tool events | The service term. Pod logs only for the life of the job | At the end of the service, a signed DPA provides return or deletion within 30 days. The managed cloud runs on DigitalOcean in London (LON1). See the [subprocessor list](https://thriveondev.com/legal/subprocessors). ## Sign-in and access People sign in to the workspace through ThriveOnDev's identity service with OIDC. Roles come from your workspace membership. See [Team and roles](https://thriveondev.com/docs/workspace/team-and-roles/). Each run is tied to the person who gave the task and their own AI access, so there is no shared agent account. ## Assurance status - A DPA draft is available for legal review. - No penetration test has been carried out yet. - ThriveOnDev is not SOC 2 certified, and no audit is in progress. For the full overview, see the [security page](https://thriveondev.com/security). Send security questions to . ## Next steps - [Deployment options](https://thriveondev.com/docs/security/deployment-options/) - [GitHub integration](https://thriveondev.com/docs/integrations/github/) --- Source: https://thriveondev.com/docs/security/data-and-isolation/ Last updated: 2026-11-20 --- # Deployment options > Use ThriveOnDev as a managed cloud service, or plan an installation in your own Kubernetes cluster delivered as a paid project with our team. ThriveOnDev runs as a managed cloud service. The platform is also built to be installed in your own Kubernetes cluster, and the first installation is delivered as a paid project with our team. ## Managed cloud This is how ThriveOnDev works by default. ThriveOnDev operates the platform. You connect Linear and GitHub and add your AI access, and runs happen in an environment reserved for your workspace. The managed cloud runs on DigitalOcean in London (LON1). Use it when you want to start quickly and don't need code or run data to stay inside your own infrastructure. See [Data handling and isolation](https://thriveondev.com/docs/security/data-and-isolation/) for what each run can access and what is kept. ## In your own cluster If your code, credentials or run history must stay inside your infrastructure, ThriveOnDev can run in your Kubernetes cluster. The platform was designed for this: each workspace gets its own namespace, network rules and per-run credentials, just as in the managed cloud. An installation in your cluster is not a self-service download. The first one is a paid project in which our team installs and verifies the platform with yours. To discuss it, write to or [book a call](https://cal.com/ivanzotov/pilot-intro). ## Choosing between them | | Managed cloud | Your cluster | | ----------------------- | --------------------------------- | ----------------------------------------------- | | Who operates it | ThriveOnDev | Your team, with ours for the first installation | | Where code is processed | ThriveOnDev's managed environment | Your infrastructure | | How to start | Connect Linear and GitHub | A scoped installation project | ## Next steps - [Data handling and isolation](https://thriveondev.com/docs/security/data-and-isolation/) - [Frequently asked questions](https://thriveondev.com/docs/reference/faq/) --- Source: https://thriveondev.com/docs/security/deployment-options/ Last updated: 2026-11-20 --- # Linear commands and labels > Every way to control ThriveOnDev from Linear, including assignment, mentions, stage commands, aliases, the Automatable label and stop. You control ThriveOnDev from Linear in five ways: assign or mention the agent, write a stage command in a comment, name an [alias](https://thriveondev.com/docs/reference/glossary/#alias), add the `Automatable` label, or press stop. ## Start work | What you do in Linear | What happens | | ----------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ | | Assign or delegate the issue to the ThriveOnDev agent | The agent replies "Starting to work on this issue…" and starts the implement stage. | | Mention the agent in a comment | Same as assigning. Use it on an issue that is already assigned to someone. | | Add the `Automatable` label | If **Automatic tasks** is on, ThriveOnDev picks up the unstarted issue within about a minute. Issues with unfinished blocking issues wait. | ## Stage commands A comment line that starts with a stage command runs that stage. You can put a mention or an alias before it. | Command | Runs | | ------------- | ------------------------------------------------------------- | | `/implement` | The implement stage: make the change and open a pull request. | | `/review` | A review of the current pull request. | | `/fix-review` | A fix of the latest review findings or failed checks. | | `/release` | The release stage: check CI and merge. | For example, this comment runs the implement stage with Claude Code on Opus: ```text @thriveondev #opus /implement ``` When a task starts from assignment or from the label, the agent writes its own `/implement` comment and runs from it, so the history shows exactly what started each run. ## Choose the agent Write an alias anywhere in the issue or comment, such as `#opus`, `#fable`, `#astra` or `#sol`. The first `#word` that matches an alias wins. Without one, the workspace's developer agent is used. > **Warning:** Any `#word` that matches an alias selects that agent, even inside a sentence. Avoid words like `#sol` in prose unless you mean the agent. ## Point at a repository If your workspace allows several repositories, or the issue does not make the target clear, add two lines to the issue description: ```text Repo: your-org/your-repo Base branch: main ``` ## Stop a run Press the stop button on the agent session in Linear. To remove a task that has not started yet, cancel it in the **To do** list of the [Workflow board](https://thriveondev.com/docs/workspace/workflow-board/). ## Replies you may see | Reply | Meaning | | ----------------------------------------------------------------- | ----------------------------------------------------------------------------- | | "Starting to work on this issue…" | The task started. | | "⏳ **Queued** — …" and the `queued` label | All slots are busy or the task waits for something. It starts by itself. | | "The agent is off for this workspace…" | The agent has not been turned on for your workspace yet. Contact ThriveOnDev. | | "No repositories have been enabled for agents in this workspace…" | An owner or admin needs to choose repositories in Settings. | | "**Automatic fix-review paused.**" | The review loop reached its limit. A person decides what happens next. | See [Troubleshooting](https://thriveondev.com/docs/reference/troubleshooting/) for the full list of problems and fixes. ## Next steps - [Start a task from Linear](https://thriveondev.com/docs/work/start-a-task/) - [Choose an agent with aliases](https://thriveondev.com/docs/agents/aliases/) --- Source: https://thriveondev.com/docs/reference/linear-commands/ Last updated: 2026-11-20 --- # Glossary > Definitions of the terms used across the ThriveOnDev docs, from task and run to alias, review loop, AI access and slot. These are the words ThriveOnDev uses in Linear, in your workspace and in these docs. Each entry stands on its own, so you can link to it. ### Agent An AI coding agent, such as Claude Code or Codex, that ThriveOnDev starts to work on a task. Each start is a separate [run](https://thriveondev.com/docs/reference/glossary/#run) in its own isolated environment. ### AI access Your personal sign-in to an AI provider, added under **Settings → AI models**. A run uses the AI access of the person who gave the task. See [Add your AI access](https://thriveondev.com/docs/agents/ai-access/). ### Alias A short name for an agent and model, written with `#`, such as `#opus` or `#sol`. Writing an alias in an issue or comment chooses the agent for that task. See [Choose an agent with aliases](https://thriveondev.com/docs/agents/aliases/). ### Automatable A Linear label. ThriveOnDev picks up unstarted issues with this label automatically when **Automatic tasks** is on for your workspace. ### Developer agent The alias your workspace uses to implement tasks when an issue does not name one. Owners and admins set it under **Settings → AI models → Developer**. ### Fix review The stage that addresses review findings or failed release checks. The agent updates the same pull request, which then goes back to [review](https://thriveondev.com/docs/reference/glossary/#review). ### Human approval An optional workspace setting. When it is on, a reviewed pull request waits in **In Human Review** until someone approves it in GitHub. See [Review loop and human approval](https://thriveondev.com/docs/work/review-and-approval/). ### Queue The list of tasks waiting to start. A task waits when every [slot](https://thriveondev.com/docs/reference/glossary/#slot) is busy, or until blocking issues, CI checks or a provider are ready. It starts by itself when the wait is over. See [Queue and capacity](https://thriveondev.com/docs/work/queue-and-capacity/). ### Release The last stage. ThriveOnDev checks that the reviewed pull request passes CI, squash-merges it, and moves the issue to **Done**. ### Review The stage in which a separate run checks the pull request against the issue. It uses the [reviewer agent](https://thriveondev.com/docs/reference/glossary/#reviewer-agent). ### Review loop Review and fix review repeating until the review approves. After three automatic fix-review rounds the loop pauses and leaves the decision to a person. ### Reviewer agent The alias that reviews pull requests. Owners and admins set it under **Settings → AI models → Reviewer**. It can be a different agent from the developer agent, or the same. ### Run One start of an agent for one stage of a task, such as implement or review. A task usually has several runs. ### Slot Room for one running agent. Your workspace runs a limited number of agents at once, and each running agent takes one slot. ### Task One Linear issue that you handed to ThriveOnDev, together with all of its runs and its pull request. ### Workflow board The **Workflow** screen of your workspace. It shows every task the agent has, in the columns of your Linear statuses. See [The Workflow board](https://thriveondev.com/docs/workspace/workflow-board/). ### Workspace Your organization in ThriveOnDev: its connections to Linear and GitHub, its people, and its settings. You open it at `thriveondev.com/orchestration`. ## Next steps - [Core concepts](https://thriveondev.com/docs/get-started/core-concepts/) - [Linear commands and labels](https://thriveondev.com/docs/reference/linear-commands/) --- Source: https://thriveondev.com/docs/reference/glossary/ Last updated: 2026-11-20 --- # Frequently asked questions > Short answers about what ThriveOnDev does, which agents it runs, who pays for AI usage, what people still decide, and where your code runs. Quick answers to the questions teams ask most before and during their first tasks. Each answer links to the page with the details. ### What does ThriveOnDev do? It runs AI coding agents on your Linear issues. An agent implements the change and opens a GitHub pull request, a separate run reviews it, and an approved pull request is merged after its CI checks pass. See [How a run works](https://thriveondev.com/docs/get-started/how-a-run-works/). ### Which coding agents can I use? Claude Code and Codex. You choose them per task with aliases such as `#opus`, `#fable`, `#astra` and `#sol`, or set defaults for your workspace. See [Supported coding agents](https://thriveondev.com/docs/agents/supported-agents/). ### Who pays for the AI usage? Each person adds their own AI access, a Claude key or a ChatGPT sign-in for Codex, and runs use the access of the person who gave the task. ThriveOnDev does not add a markup to model usage. See [Add your AI access](https://thriveondev.com/docs/agents/ai-access/). ### Can the agent merge without a person? Yes, by default an approved pull request is merged after CI passes. Owners and admins can turn on human approval, so every pull request waits for an approving GitHub review first. See [Review loop and human approval](https://thriveondev.com/docs/work/review-and-approval/). ### What happens if the review keeps rejecting the change? After three automatic fix-review rounds the agent stops and leaves a comment in Linear. A person then clarifies the issue or asks for another round with `/fix-review`. ### How many tasks can run at once? Your workspace runs a limited number of agents at the same time. Extra tasks wait in the queue and start by themselves when a slot frees up. See [Queue and capacity](https://thriveondev.com/docs/work/queue-and-capacity/). ### Which repositories can the agent change? Only the repositories and branches that an owner or admin chose in Settings. Installing the GitHub App does not allow any repository by itself. See [GitHub integration](https://thriveondev.com/docs/integrations/github/). ### Where does my code run? Each run is a separate job in an environment reserved for your workspace, with network rules and short-lived credentials. See [Data handling and isolation](https://thriveondev.com/docs/security/data-and-isolation/). ### Can I run ThriveOnDev in my own cluster? The platform is built for installation in your own Kubernetes cluster. The first installation is delivered as a paid project with our team. See [Deployment options](https://thriveondev.com/docs/security/deployment-options/). ## Next steps - [What is ThriveOnDev](https://thriveondev.com/docs/get-started/what-is-thriveondev/) - [Troubleshooting](https://thriveondev.com/docs/reference/troubleshooting/) --- Source: https://thriveondev.com/docs/reference/faq/ Last updated: 2026-11-20 --- # Troubleshooting > Fix the common problems with ThriveOnDev tasks, such as an agent that does not start, AI access errors, long queues, review limits and failing CI. When a task does not go as expected, the agent says why in the Linear thread, and the task card on the Workflow board shows a stop reason with **What is needed**. Find the message or reason below. ## The agent does not respond - **"The agent is off for this workspace."** The agent has not been turned on for your workspace yet. Contact ThriveOnDev at . The Workflow board shows **Agent off** in this case. - **No reply at all.** Check that the issue is in a Linear team your workspace works in (**Settings → Linear → Choose teams**) and that you assigned, delegated or mentioned the ThriveOnDev agent, not a person. - **"Agent access denied"** or **"Message was not accepted"**. This usually means you don't have permission to give tasks yet. Ask an owner or admin to turn on **Can give tasks** for you. See [Team and roles](https://thriveondev.com/docs/workspace/team-and-roles/). ## The agent couldn't be started Most start problems are setup problems. Open the task card and follow **Open Settings**: | Stop reason | What is needed | | --------------------------- | ------------------------------------------------------------------------------------------------------------------------ | | **AI access not connected** | Add AI access for the agent this task uses. See [Add your AI access](https://thriveondev.com/docs/agents/ai-access/). | | **AI access rejected** | Remove your access and add it again. | | **Repository not allowed** | An owner or admin adds the repository under **Settings → GitHub → Choose repositories**. | | **Setup incomplete** | Finish the station Settings marks. | | **Model not available** | Choose another alias, or ask an owner or admin to change the developer or reviewer agent. | | **Agent didn't start** | Start the task again. If the agent says "Contact support to check the startup failure", write to . | If the agent asks for `Repo: owner/name` and `Base branch: branch`, the issue does not point to exactly one allowed repository. Add those two lines to the issue description. ## The task waits in the queue for a long time The `queued` comment in Linear names the reason and the position. Usually every slot in your workspace is busy. Tasks start by themselves as others finish. To free a slot, cancel tasks you no longer need in the **To do** list. See [Queue and capacity](https://thriveondev.com/docs/work/queue-and-capacity/). ## The review keeps rejecting the change After three automatic fix-review rounds, the agent posts "**Automatic fix-review paused.**" and stops. The issue and the review disagree, or the task is too big. Clarify the acceptance criteria in the issue, or split it, then comment `/fix-review`. Or fix the pull request yourself and comment `/review`. See [Review loop and human approval](https://thriveondev.com/docs/work/review-and-approval/). ## CI fails or the pull request conflicts At release, failing CI or a conflict with the base branch sends the task back to fix review. The agent fixes the problem and merges the base branch in when needed, up to three times per task. If checks are still running, the task waits and continues by itself. If it keeps failing, fix the cause or re-run flaky checks in GitHub, then comment `/release` or `/fix-review`. ## AI usage runs out - **AI credits used up** or **AI usage limit reached**: the provider account behind the AI access has no usage left. Top it up or wait for the limit to reset, then start the task again. - **"Access needs to be renewed"**: add your AI access again. For Codex, paste a fresh `auth.json`. - **AI provider unavailable**: the provider is down. The task waits and continues by itself. ## The task reached its time limit A run stops after 5 hours. Split the issue into smaller ones that each fit in one pull request. See [Write issues an agent can finish](https://thriveondev.com/docs/work/write-good-issues/). ## The task is too large **Task too large** means the change does not fit in one run. Split it into several issues and connect them as blocking issues in Linear. ## Still stuck Write to with the Linear issue link, or check the [system status](https://status.thriveondev.com/). ## Next steps - [Follow a run](https://thriveondev.com/docs/work/follow-a-run/) - [Frequently asked questions](https://thriveondev.com/docs/reference/faq/) --- Source: https://thriveondev.com/docs/reference/troubleshooting/ Last updated: 2026-11-20