Continue Agent Run
Advance a persisted agent run from its current state. Dispatches on the body shape and the persisted run state (see ADR-003 in specs/agno/features/checkpointing/decisions.md).
Variants:
- PAUSED + tools provided → apply HITL tool results, resume
- PAUSED + resolved admin approval (empty tools) → apply resolution, resume
- RUNNING / ERROR (no unresolved HITL requirements) → resume from last persisted state
- COMPLETED + new tools → continue with appended messages
Tools Parameter: JSON string containing array of tool execution objects with results. Optional — only required when the persisted run has unresolved HITL requirements.
Authorizations
Bearer authentication header of the form Bearer <token>, where <token> is your auth token.
Body
JSON string of tool call results to continue the paused run
Optional new user-message text to append to the run before resuming. Use for continuing a COMPLETED run with a follow-up, or adding context to a RUNNING/ERROR resume.
Continuation boundary. Use 'end', 'last_user', or a numeric message index.
When true, clone the run with a new run_id before resuming. The original is untouched; the clone becomes a sibling within the same session, with forked_from_run_id set.
Sugar: regenerate the last response of this run. Auto-computes continue_from='last_user' to land just after the last user message. Pair with additional_instructions to steer the new output. By default the original response is hidden from history (replaced); pass replace_original=false to keep both the original and the regenerated response visible side by side.
Only valid with regenerate=true. Controls history visibility of the original response; the original run is always retained in storage. Defaults to true: the original is marked REGENERATED and hidden from history so the new response replaces it. Pass false to keep both the original and regenerated responses visible.
Only valid with regenerate=true: extra guidance appended as a user message before re-generation. Friendly alias for input.
Session ID for the paused run
User identifier for tracking and personalization
Enable streaming responses via Server-Sent Events (SSE)
Run continue in background (survives client disconnect). Requires database. Use /resume to reconnect.
Response
A non-streaming run response or a server-sent event stream. Streaming execution failures are delivered as events after the stream starts.
Unique identifier for the run
Parent run ID if this is a nested run
Agent ID that executed this run
User ID associated with the run
Run status (PENDING, RUNNING, COMPLETED, ERROR, etc.)
Input provided to the run
Output content from the run
Format of the response (text/json)
Reasoning content if reasoning was enabled
List of reasoning steps
Performance and usage metrics
Message history for the run
Tools used in the run
Events generated during the run
Run creation timestamp
References cited in the run
Citations from the model (e.g., from Gemini grounding/search)
Reasoning process messages
Session state at the end of the run
Images included in the run
Videos included in the run
Audio files included in the run
Files included in the run
Audio response if generated
Input media attachments
Followup suggestions generated after the run
If this run was forked from another run, the source run's ID
If this run was forked, the message index at which the source was truncated
If this run was created via session branch, the source session's ID
If this run was produced via regenerate=true, the source run's ID
Message index of the most recent mid-run checkpoint (checkpoint='tool-batch' runs)