Activity & observability

Every time a workflow runs, Terse records what happened: which trigger fired, what the model decided, which tools were called, whether it succeeded or failed, and how long it took. This data powers the Activity tab, the Stats dashboard, notifications, and the weekly self-improvement reviews. You don’t need to add logging to your code. Observability is built into the platform.

​

Run lifecycle

Each run moves through a defined set of statuses:

​

What gets recorded

Every run produces a run record with three layers of detail:

​

Run summary

The top-level record captures the basics: status, trigger source, trigger title, the model’s decision (processed or skipped), decision reasoning, whether it was manually triggered, and timestamps. This is what you see in the Activity list.

​

Action trace

Each side effect (a CRM record updated, a Slack message sent, a query executed) is logged as an action with:

• The integration and tool name
• Whether it was a read or write
• A human-readable summary
• An optional link to the result (e.g., a URL to the created record)

Actions are categorized by type: create, update, delete, read, approve, or error.

​

Raw event stream

The full model conversation is stored as a sequence of raw events: every prompt, tool call, tool result, and model response, in order. This is what powers the run replay in the Activity drawer. You can step through the entire execution and see exactly what the model was thinking and doing at each point.

​

Activity tab

The Activity tab shows all runs across your workflows in one place, with the most recent runs first. Search across run IDs, trigger titles, event content, decision reasons, and workflow names. Filter by status to focus on failures, in-progress runs, or runs awaiting approval. Skipped runs are hidden by default to reduce noise. Toggle them on when you need them. Filter by date range to narrow down to a specific time window. Inspect a run by clicking any row. This opens a drawer with the full run replay: the trigger payload, the model conversation, tool calls with their arguments and results, and the final output. For runs awaiting approval, you can approve or reject directly from here. From the run drawer or list, you can re-trigger the workflow with the same trigger payload (a new run on Terse). To execute that payload against your local code instead, use terse replay with the run ID. To list past runs and optional trigger payloads from the terminal, use terse history.

​

Stats dashboard

The Stats page aggregates run data into a dashboard with configurable time intervals (1 hour to 1 year):

​

Notifications

Terse can notify you about run outcomes through multiple channels: Notifications are configured per-workflow. You choose which action types to notify on (creates, updates, deletes, errors, and approval requests) and Terse delivers to your preferred channel. Slack is used when you have an active Slack destination configured; otherwise, notifications fall back to email. Every notification sent is recorded in your notification history, so you have an audit trail of what was delivered and when.

​

How observability feeds self-improvement

The same run data that powers Activity and Stats is also what the self-improvement system uses. The weekly review pulls run summaries, failed run traces, and action logs to identify patterns and recommend changes. Better observability means better improvement suggestions.

​

Where to go next