Runs
Every piece of agent work is a run. Find runs on Agents → Runs, read exactly what one did on its Run Detail page, and resume or dismiss the ones that stopped.
A run is one piece of work an agent did: building strategies from an idea, a validation pass, a post-mortem, a routine firing, an approval's troubleshoot. Each run has a status, an agent and model, its cost, and a full record of what it did — prompt, model turns and every tool call. When something stalls or behaves oddly, this is where you read exactly what happened.
Older builds called these tasks and listed them on a separate Tasks page. That page is gone: /tasks now redirects to Agents → Runs, keeping any filters in the link.
Find a run: Agents → Runs
Open Agents → Runs. Runs are grouped into buckets:
| Bucket | What's in it |
|---|---|
| Needs a look | Blocked and failed runs — the ones waiting on you |
| Blocked | Stopped partway; can be resumed from their checkpoint |
| Failed | Ended with an error |
| Running | Working now |
| Queued | Waiting to start |
| Done | Finished |
| Cancelled | Cancelled or rejected |
| All | Everything |
Narrow it with All agents or the search box (Search title, strategy, model, error…). Click a run to see its agent, kind, strategy, timing, model, usage and retries, What it was asked to do, and — if it stopped — why.
Resume or dismiss
- Resume appears on blocked runs. The run continues from its saved checkpoint and spends on your model again.
- Dismiss clears a blocked, failed or cancelled run from Needs you and every attention count. The run and its transcript stay in the history.
Tick several rows to Resume N or Dismiss N in one go; you're asked to confirm.
Read a run: Run Detail
Open transcript opens the run's Run Detail page (/tasks/<id>). The header shows the status, cost, tokens and model, and flags a truncated record. Refresh reloads it; ← Back returns to where you came from.
| Tab | What it shows |
|---|---|
| Overview | Display and internal IDs, title, agent, priority, strategy, stage, when it was created, started and completed, how long it took, and any error or scheduled retry. |
| Transcript | The run turn by turn: the prompt, each model reply (with reasoning where the provider returns it), every tool call with its arguments and result, and any nudge the runner added. It's written as the run goes, so you can read an agent's work while it's still running. |
| Audit Log | The lifecycle timeline: created, started, finished or errored, with timestamps. |
| Tool Calls | Each tool the run called, in order, with input, output, duration and any error. |
| Run Data | The run's input data and output data as JSON. |
| Raw JSON | The complete stored record. |
Over the API, the same record is GET /api/pipeline/tasks/<display_id>/audit.
Steps: find out why a run failed
Find it
On Agents → Runs, open Needs a look (or Failed) and click the run.
Read the error
Click Open transcript. The Overview tab shows the error and whether a retry is scheduled.
Find the failing call
Tool Calls shows the exact call that broke, with its input and error. Transcript shows what the agent was thinking when it made it.
Decide
Resume it if the cause is fixed (a provider back up, data now downloaded), or dismiss it so it stops counting as needing you.
Interrupted runs
A run cut off by a restart is checkpointed, not lost, and listed under Resumable Runs on Diagnostics. Most are safe to resume there. Two kinds that act on the outside world — trade_execution and phantom_repair — are never resumed automatically and show Resume (caution), because re-running them could repeat that effect. Check the run's Tool Calls first.
Good to know
- Stages change only through promotions and demotions. A run can't quietly rewrite a strategy's stage; out-of-band stage writes are blocked.
- Costs are estimates from token counts and list prices. Your provider's dashboard is the bill.
Forven is research software. A run records what the system did, not what markets will do, and nothing here is financial advice.
Related
- Agents — who does the work, and the rest of the Agents page
- Diagnostics — resumable runs after a restart
- Scheduled routines — prompts that create a run each time they fire
- Troubleshooting
Scheduler & jobs
The Forven scheduler — the 47 built-in background jobs that drive strategy creation, testing, trading, data and maintenance; where to see them on Agents → Schedules, what each system mode lets run, and how to pause or re-time them.
The assistant & deep-dive
The in-app assistant is a page-aware chat that can explain, navigate, and — if you allow it — create strategies and run backtests, with confirmation cards for anything consequential. Deep-dive focuses it on one strategy.