Skills

Base URL https://api.tilion.dev. Every request needs Authorization: Bearer $TILION_API_KEY.

POST /v1/task

State a goal in natural language against a domain. Tilion replays a cached skill when it has one and derives a new one when it does not, so repeat calls get cheaper.

Request

Field	Type	Notes
`domain`	string	Required. For example `example.com`
`goal`	string	Required. For example “get the page heading”
`session_id`	string	Required
`url`	string	Page to navigate for a cold derive
`schema_keys`	string[]	Expected output field names
`variables`	object	Values for `{slot}` placeholders
`binding`, `wait_for`	object	Advanced overrides

Response

{
  "data": {}, "decision": "REPLAY", "success": true, "from_cache": true,
  "self_healed": false, "tokens_used": 0, "tokens_saved": 4120,
  "skill_id": "...", "reason": "...", "vertical": "retail", "warm_start": true
}

decision is REPLAY, DERIVE, or HEAL.

List your skills

GET /v1/skills

Cache statistics

GET /v1/determinism/stats

The endpoint reports cache-hit rates and token savings across your skills, scoped to your account. Use it to see how much repeat work is served from cache rather than re-derived.

Get started

MCP

Sessions

Fetch & jobs

Content

Agent & tasks

Live view & recordings

Account

Reference

Request

Response

List your skills

Cache statistics

​Request

​Response

​List your skills

​Cache statistics

Request

Response

List your skills

Cache statistics