AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.
Conductor Create Session
Starts a new agent session (chat) in an existing Conductor workspace, optionally with a first prompt, and optionally waits for the reply. Requires your own Conductor API key, created at app.conductor.build/users/api-keys and added through AutoGPT’s credentials UI. Select that credential for each Conductor block; no server-wide default key is used.
Conductor Create Session
What it is
Start a new agent session (chat) in an existing Conductor workspace, optionally with a first prompt, and optionally wait for the agent's reply.
How it works
The block posts to POST /v0/sessions with workspaceId, agent and any non-blank model, effort, name or message. When message is given the returned initial_message_id identifies the prompt and next_after is the prompt's transcript row ID to pass as after to Get Session (the receipt itself while the prompt is still queued, which Get Session also accepts); with wait_for_reply the block waits the same way as Send Message: it reads the transcript and then the session status on every poll (a wall-clock bound of timeout_seconds, reported through timed_out), correlates the rows with the prompt's turn, treats the session as done only once it is idle after the turn progressed past its startup events, and returns the visible agent text joined into reply and the kept-row message_count (truncated is set when the turn exceeded the 1000 rows kept or its start was older than the rows read). The raw rows are only returned, in messages, when include_messages is on.
Inputs
| Input | Description | Type | Required |
|---|---|---|---|
| workspace_id | Workspace to add the session to | str | Yes |
| message | Initial prompt for the agent; leave empty for an idle session | str | No |
| agent | Agent to run | "claude" | "codex" | "cursor" | "acp" | No |
| model | Model id such as fable-5-1, opus-5-5-1m, sonnet-5-1m, gpt-6-astra or auto. Leave empty for Conductor's default. | str | No |
| effort | Reasoning effort; leave empty for the default | "" | "none" | "low" | "medium" | "high" | "xhigh" | "max" | "ultra" | No |
| fast_mode | Enable fast mode | bool | No |
| name | Session name | str | No |
| wait_for_reply | After sending the initial prompt, wait until the agent is idle and return its reply | bool | No |
| timeout_seconds | How long to wait for the reply | int | No |
| poll_interval_seconds | Seconds between status checks while waiting | int | No |
| include_messages | Also return the raw transcript rows of the turn in messages. Off by default: a long turn is hundreds of kilobytes, while reply, session_status and next_after cover the usual needs | bool | No |
Outputs
| Output | Description | Type |
|---|---|---|
| error | Error message if the operation failed | str |
| session_id | ID of the new session | str |
| deep_link | Link that opens the session | str |
| initial_message_id | ID of the initial prompt message, empty when none was sent | str |
| next_after | Transcript row ID of the prompt's row; pass it as after to Get Session to read the agent's turn. Falls back to initial_message_id while the prompt has no row yet, which Get Session also accepts |
str |
| session_status | idle, working or error once waiting finished | str |
| reply | Text the agent produced in response | str |
| message_count | Number of transcript rows of the turn that were kept | int |
| messages | Raw transcript rows of the turn, oldest first; only emitted when include_messages is on | List[Dict[str, Any]] |
| timed_out | True when the wait ended before the agent went idle | bool |
| truncated | True when the turn produced more rows than are kept; the newest ones were kept and reply may be incomplete | bool |
| error_message | Session error, if any | str |
Possible use case
Second opinion: Open a Codex session in a workspace where Claude already worked and ask it to review the diff.
Split work: Start one session per sub-task in the same workspace so they share the checkout.