Executions

An execution is the saved record of one run: an AI agent working through a task, or a workflow running from its trigger to its last node. Executions answer the two questions you have after anything automated happens: did it run, and did it work.

You find them under Executions in the Activity group of the Console sidebar. The page itself is titled Runs because it covers agents and workflows in one list.

What an execution records

Traces

Every execution carries an ordered list of traces, one per step. A trace records:

Read the traces top to bottom to follow the agent’s reasoning: which tool it chose, the arguments it sent, what came back and what it decided next. When a run fails, the last trace with an error usually tells you why.

States

The State filter on the Runs page offers these values:

Runs recorded by the agent runtime use a finer set of end states, which you may see on AI Agent runs:

Find a run

Runs are scoped to the project you have selected. Switch project to see another project’s runs.

Stop or delete runs

These actions need the administrator or owner role, and none of them can be undone.

Delete matching is not offered for workflow runs. Delete those one at a time from the row actions.

If a bulk delete reports that nothing was deleted, your role in the organization is probably not high enough.

Debugging a failed run

1. Start from the end. Open the Error tab, then the last few traces. Most failures are a tool that returned an error or a model step that ran out of room.
2. Check the input. A surprising answer often comes from a surprising input. Compare the Input tab with what you expected the channel or trigger to send.
3. Look at the tool calls. Check the arguments the agent sent. Wrong arguments usually mean the tool’s description or the agent’s instructions are unclear.
4. Compare with a good run. Filter by the same workflow or session and open a run that worked.
5. Watch it live. Reproduce the problem with the agent’s Live activity panel open (see Sessions) to see each step as it happens.
6. Change one thing at a time. After editing instructions, skills or tools, run the same input again and compare the traces.

Executions API

The Executions API gives API keys read access to the same run records the Console shows: AI agent runs with their traces, inputs, outputs, errors and cost. Use it to pull run history into your own monitoring, alert on failures, or attach a run’s traces to a support ticket automatically.

It covers AI agent runs. Workflow runs stay in the Console for now.

Endpoints

Base URL https://app.syntic.ai/v1 (the paths above already include /v1). Authenticate with a private key and send syntic-beta: agents-2026-11-01 while in preview. Reads work with any key in the workspace. Terminate and delete follow the Console rule: the key’s creator must be an Administrator or Owner, otherwise 403 role_required.

List runs

GET /v1/agents/executions

The list returns run summaries without traces:

{
  "data": [
    {
      "id": "exec_01JC7N0Q8S2U4W6Y8A0C",
      "type": "agent_execution",
      "agent_id": "agt_01JC4R8W6Q2N9X3T5V7Y",
      "agent_name": "refund-desk",
      "agent_version": 4,
      "session_id": "sesn_01JC7M2K4P6R8T0V2X4Z",
      "state": "error",
      "started_at": "2026-11-05T10:00:01Z",
      "ended_at": "2026-11-05T10:00:19Z",
      "duration_ms": 18234,
      "error": "Tool lookup_order returned an error: order not found",
      "usage": { "input_tokens": 9120, "cached_input_tokens": 6400, "output_tokens": 412, "cost_usd": "0.0294" },
      "trace_count": 7
    }
  ],
  "has_more": true,
  "first_id": "exec_01JC7N0Q8S2U4W6Y8A0C",
  "last_id": "exec_01JC7MZ7R1T3V5X7Z9B1"
}

Read one run

GET /v1/agents/executions/{execution_id} returns the summary fields plus input, output and traces. Each trace is one step:

Runs with more than 200 traces return the first 200 and "traces_truncated": true; page through the rest with /traces?after_step=199.

# Every failed run in the last day
curl -sS -G https://app.syntic.ai/v1/agents/executions \
  -H "Authorization: Bearer $SYNTIC_API_KEY" \
  -H "syntic-beta: agents-2026-11-01" \
  --data-urlencode "state=error" \
  --data-urlencode "state=timeout" \
  --data-urlencode "started_after=2026-11-04T10:00:00Z"

const params = new URLSearchParams([
  ['state', 'error'], ['state', 'timeout'],
  ['started_after', new Date(Date.now() - 86_400_000).toISOString()],
])
const res = await fetch(`https://app.syntic.ai/v1/agents/executions?${params}`, {
  headers: {
    Authorization: `Bearer ${process.env.SYNTIC_API_KEY}`,
    'syntic-beta': 'agents-2026-11-01',
  },
})
const { data } = await res.json()
for (const run of data) console.log(run.id, run.agent_name, run.error)

import os, requests
from datetime import datetime, timedelta, timezone
 
since = (datetime.now(timezone.utc) - timedelta(days=1)).isoformat()
resp = requests.get(
    "https://app.syntic.ai/v1/agents/executions",
    headers={"Authorization": f"Bearer {os.environ['SYNTIC_API_KEY']}", "syntic-beta": "agents-2026-11-01"},
    params=[("state", "error"), ("state", "timeout"), ("started_after", since)],
)
resp.raise_for_status()
for run in resp.json()["data"]:
    print(run["id"], run["agent_name"], run["error"])

Terminate and delete

• POST /v1/agents/executions/{id}/terminate ends a run in in_progress with state terminated and returns the updated summary. On any other state: 409 execution_not_running.
• DELETE /v1/agents/executions/{id} removes the run and its traces permanently and returns {"id": "...", "type": "agent_execution_deleted"}. Bulk deletion stays in the Console.

Both actions write an audit event with the key id as the actor.

Errors

Errors use the error envelope.

Limits

Reading executions is free.

Build notes for the platform team

• This is the quick win from the roadmap: add an API-key auth path to the existing agent-executions function’s list, get, terminate and delete actions, scoped to the key’s workspace, and switch offset paging to cursors.
• Map the Console’s workflow-run state names only in the Console; the API uses the agent-run states listed above.

Related

• Sessions: the conversations that trigger runs.
• Tools & skills: what an agent may do during a run, and approvals.
• Audit logs: who changed what in the organization.