Models

Every task runs on a coding agent and a model. In the app both come from the agent you assign the work to, so there is no per-task model choice. Over the API, set the coding agent, model, and effort on a task create that carries a prompt, and they apply to the session that create starts.

Choosing a model

The model belongs to the agent. Open the agent from Team. On a phone, choose Model; on web and wider layouts, open Configure and select Model. The picker groups models under the coding agent that runs them:

• Claude Code (Anthropic)
• Codex (OpenAI)

Pick a coding agent, then a model within it. Leave it on Default to follow the account, which is Opus 5.5 under Claude Code. The choice applies to the tasks that agent starts from then on; a session already running keeps the model it started with. To run the same work on a different model, assign it to an agent set to that model.

Claude Code (Anthropic)

Context Window

Every Claude model works against a 200,000-token context window, and a session that outgrows that window cannot take another turn.

This shows up on long-running work: a task that reads a lot of a large repository, or that runs for hours across many turns, can reach the cap. When it does, start a fresh task for the work that is left, or narrow what the agent has to read.

Codex (OpenAI)

The app bundles this curated Codex list so the picker is ready on first paint and remains usable if the model catalog cannot load. After GET /v1/models loads, its catalog becomes authoritative.

Agents on Default, including existing agents, use the current default for new sessions. An explicit model selection stays pinned, and a session already running keeps its selected model.

An unknown or unavailable model falls back to the agent's default rather than failing the task, so the session always runs on a supported model.

Reasoning effort

Effort is how much thinking a turn spends before it answers. Each coding agent has its own levels and its own default:

Set it on the agent: open the agent, go to Settings, and use the Effort control beside Model. The levels shown are the ones the agent's coding agent accepts, and an agent whose coding agent has no effort levels shows no control.

A finished turn records the effort it ran at. The turn's details show it beside the model. When no effort was pinned, the app shows the coding agent's default when the served catalog can identify one; a coding agent with no effort axis shows none.

Where the effort comes from

For each run, in order: the effort passed on the request, then the effort set on the agent, then the coding agent's default. The agent's setting only carries over when the run resolves to the same coding agent the setting was made on, because the levels are per coding agent; a run that switches coding agents ignores it and takes the new one's default. A level that is not in the chosen coding agent's vocabulary falls back to that default rather than failing the task.

Choosing a model over the API

Set the coding agent, model, and effort on task create with the harness, model, and effort fields. All three apply to the session that create starts when the request carries a prompt; a create without one starts no session and drops them. None of the three is stored on the task, and creating a session under an existing task takes none of them: that session resolves from the acting agent's current settings. effort applies to that one run and leaves the agent's own setting alone. GET /v1/models reports the models on offer and, per coding agent, the efforts it accepts and its default. See Tasks and Sessions.