Skip to content

Local web API

ATLAS exposes a small localhost HTTP API for the runtime dashboard. It is meant for local monitoring, notebooks, benchmark harnesses, and lightweight external dashboards.

The API is read-only except for the managed-dashboard shutdown endpoint.

Start the server

atlas-dashboard \
  --trace-output ./atlas-program \
  --store-dir ~/.atlas-skill/taxonomies

By default the dashboard binds to 127.0.0.1:8765.

Integrations may start the dashboard automatically when dashboard is true in atlas.json.

GET /api/health

Returns whether the dashboard process is serving the expected program.

Example:

{
  "program_id": "program-123",
  "status": "ok"
}

Use this for lightweight liveness checks.

GET /api/taxonomy

Returns the current taxonomy view for one program, overlaid with program-local runtime evidence.

Example shape:

{
  "program_id": "program-123",
  "taxonomy_id": "tax-20260708T000000Z-example",
  "bound_taxonomy_id": "tax-original",
  "is_latest_successor": true,
  "repo": "owner/repo",
  "domain": "display domain",
  "codes": [
    {
      "code_id": "A.1",
      "name": "Skipped verification",
      "description": "The agent declared completion without checking the result.",
      "fire_count": 2,
      "task_firings": [
        {
          "task_id": "session-0118",
          "label": "UID0118 ✗",
          "count": 2
        }
      ],
      "runtime_evidence": [
        {
          "seq": 4,
          "timestamp": 1780000000.0,
          "gate": "final_gate",
          "task_id": "session-0118",
          "task_label": "UID0118 ✗",
          "checkpoint_id": "cp-final",
          "evidence": "The final answer was submitted before any validation.",
          "correlate": "The trajectory matches A.1.",
          "decide": "Run validation before submitting."
        }
      ],
      "fields": [
        {
          "name": "category",
          "value": "Verification"
        }
      ]
    }
  ],
  "clean_checkpoints": [
    {
      "seq": 5,
      "timestamp": 1780000010.0,
      "checkpoint_id": "cp-clean",
      "gate": "final_gate",
      "task_id": "session-0120",
      "task_label": "UID0120 ✓",
      "none_apply": true,
      "considered": ["A.1"],
      "observe": "Validation was performed.",
      "correlate": "No evidence-supported failure remained.",
      "decide": "Proceed."
    }
  ]
}

Notes:

  • taxonomy_id is the latest taxonomy visible to the program.
  • bound_taxonomy_id is the taxonomy the program originally held before lineage resolution.
  • repo and domain are display metadata only.
  • runtime_evidence is program-local. It overlays the taxonomy view without mutating the stored taxonomy record.
  • clean_checkpoints are accepted reflections where no code fired.
  • Evidence text is clipped server-side for dashboard readability.

POST /api/shutdown

This endpoint exists only for dashboards started by ATLAS integrations through the managed-dashboard lifecycle. It requires the private X-ATLAS-Token header written into the program's .atlas-dashboard.json state file.

Do not call this endpoint from external dashboards unless you own the managed dashboard lifecycle.

Security model

The dashboard is designed for localhost use. Do not expose it publicly without an external authentication layer.

The API can include task labels, evidence snippets, checkpoint IDs, and reasoning captured during runtime. Treat the response as run data, not a public asset.