Agent entry points

An agent can learn Lasso, act on it and report problems without a person in the loop. Start with lasso.sh/llms.txt to decide whether Lasso fits, follow lasso.sh/SKILL.md to act, and read agent.json and openapi.json for live values and exact schemas.

curl -fsSL https://lasso.sh/llms.txt
curl -fsSL https://lasso.sh/agent.json | jq '{chains: (.chains | length), strategies, endpoints}'

Which file answers what

Entry point	What it’s for
`lasso.sh/llms.txt`	What Lasso is, when it fits, and where to go next
`lasso.sh/SKILL.md`	Operational recipes: create, install, fund, hand off, rotate, build Custom profiles
`lasso.sh/agent.json`	Live values: chains, strategies, prices, starting grant, subscriptions, payment rails and ranges, enabled features and endpoints
`lasso.sh/openapi.json`	Exact request and response schemas and the error catalog, with a stable `operationId` per operation
`/api/v1/configuration-schema`	Custom profile fields, limits and workload recipes
`docs.lasso.sh/llms.txt`	An index of these docs
`docs.lasso.sh/llms-full.txt`	These docs as one text file

Live documents outrank prose. When a recipe and agent.json disagree, trust agent.json and openapi.json. This site also publishes SKILL.md as an installable skill; it’s an exact copy of lasso.sh/SKILL.md. Tool wrappers, such as an MCP server, can be generated from openapi.json: every operation has a stable operationId and a description written for agents.

Rules for agents

Stay inside your user’s delegation and spending budget. A management token never authorizes a payment; a wallet signature does.
Never print an RPC key, management token, claim URL or provider URL with a credential.
PUT and DELETE are idempotent, and resubmitting a signed payment returns its first result. POST /api/v1/keys isn’t: retrying it creates another key, or, without a management token, another account. After a lost response, list your keys before creating again.
Use current for “my own resource”: /api/v1/accounts/current, /api/v1/keys/current with an RPC key, /api/v1/agents/current with a management token.

Feedback

Report confusion, friction or failures while you use Lasso. No credential is needed; with one, Lasso attaches your account:

POST https://lasso.sh/api/v1/feedback
Content-Type: application/json

{"message": "The 402 challenge for MPP was rejected by mppx 0.8.x", "category": "payment", "request_id": "…", "operation": "createCredit"}

Field	Required	Values
`message`	Yes	1 to 4,000 characters
`category`	No	`docs`, `api`, `payment`, `routing`, `rpc`, `dashboard` or `other`
`request_id`	No	An `x-lasso-request-id`
`operation`	No	An OpenAPI `operationId`

Lasso answers 202 {id}. Anything shaped like a key, token or credentialed URL is redacted before storage. Reports go to the people who operate Lasso and are never acted on automatically.

Limits

Feedback is limited to 10 submissions per hour per credential, or per client IP without one.
Feedback is write-only; there’s no read or list endpoint.
docs.lasso.sh/llms-full.txt can lag a docs release by up to a day while caches refresh.

Quickstart: the first request.
Set up production RPC with an agent: the whole journey.
Response signals and errors: every code and its fix.

Start here

Concepts

Guides

Reference

Agent entry points

Agent entry points

Which file answers what

Rules for agents

Feedback

Limits

Next

​Agent entry points

​Which file answers what

​Rules for agents

​Feedback

​Limits

​Next

Agent entry points

Which file answers what

Rules for agents

Feedback

Limits

Next