Talon FlowTalon Studio v1.0.2-beta

Executions

Test and run flows, then watch every step: each real run is a recorded, streamed timeline at /flow/executions.

Running & testing

In the builder, click Test run to dry-run the unsaved graph. It runs synchronously and shows each node's input and output in place; searches, notifications, agent runs, board tasks, LLM calls and propose steps are stubbed, so nothing is staged and no tokens are spent. Test runs are not recorded in Executions.

Run a saved flow for real with Run in the Automations list (you confirm first; the run uses AI tokens), or let its trigger fire it. Each node may retry 0–5 times, every attempt bounded by FLOW_NODE_TIMEOUT_MS (default 120 s), and the whole run by FLOW_RUN_TIMEOUT_MS (default 15 minutes).

Open Talon Flow → Executions (/flow/executions) to follow it.

Talon Flow → Executions

The executions list with a run in flight: it refreshes itself every 5 seconds while a run is active. Open a row to stream the per-node timeline and jump to any change requests the flow proposed.

Topological walk

Execution follows a topological order with branch-skip propagation: when an if or switch skips a branch, its downstream nodes are skipped too, not failed. The timeline shows that as "Branch not taken" on the skipped step, and a node that failed after retries records "(after N attempts)" alongside the error.

Observability

Every real run is recorded as an agent run with one step per node, holding the node's input, output (each capped at 8,000 characters of JSON) and duration. The Executions list shows the active workspace's flow runs, taken from its 50 most recent agent runs, with each run's trigger and status; it polls every 5 seconds while a run is active. Opening a run streams its timeline live over server-sent events, with links to any change requests it proposed.

Succeeded

The run finished. A run that staged at least one change request counts as succeeded even if some nodes failed.

Failed

At least one node failed and nothing was staged, or the run itself errored.

Stale runs

A run still marked running after AGENT_RUN_STALE_MS (default 30 minutes), for example after a restart, is marked failed.

Failure notifications

A failed flow run sends a flow_failure notification to the organisation's enabled notification destinations, and a run that staged change requests sends an agent-proposal notification. Each destination chooses which event types it receives.

Same table as Claw

Flow runs are not a separate log: they are rows in the same agent-runs table as Claw agent runs, so they obey the same concurrency, daily-cap and budget limits. Talon keeps run history in the database with no automatic retention window.