developers
The looprig developer portal.
Everything a client needs, at a fixed address. looprig is a set of Apache-2.0 Go modules you run yourself, so there is no API key to request and no sandbox tenant to provision. What this page gives you instead is every machine-readable surface of this site, the conventions that govern it, and the entry point for each module.
Resources
- 01
Documentation
Guides, runnable examples, product pages, and one page per public module. Every page has a raw Markdown twin at the same path with a .md suffix.
- 02
Agent index
The compact machine-readable routing index: what looprig is for, when to reach for it, and links into the Markdown corpus. Start here before crawling.
- 03
Agent index (JSON)
The same routing information as llms.txt in a typed JSON document: use cases, boundaries, resources, and the conventions of this origin, ready to branch on.
- 04
Complete corpus
Every published documentation page concatenated into one Markdown artifact, for tools that cannot fetch pages progressively.
- 05
OpenAPI specification
OpenAPI 3.1 description of this website's content API: the agent index, Markdown twins, search index, sitemap, and feed, with the negotiation, error, and rate-limit conventions that apply to them.
- 06
Source
Every module is an independent Apache License 2.0 Go repository with its own tests, tags, and README. Released versions are immutable Git tags.
- 07
Quickstart
Install the Go modules, configure a model, and run a first session end to end without cloning the development workspace.
- 08
Journal
Engineering notes on the runtime, appended in order and never rewritten. Useful for understanding why a boundary is where it is.
- 09
Contact
How to reach the maintainers, report a security issue, or ask a question that the documentation does not answer.
Conventions
- 01
content negotiation
Send Accept: text/markdown to any page URL and you get the Markdown representation of that page; send text/html, or nothing, and you get the rendered page. Quality values are honoured, and a request that accepts neither is answered 406. Every response carries Vary: Accept, Accept-Encoding.
- 02
markdown twins
Every documentation page is also published at the same path with a .md suffix, linked from the page head with rel="alternate". The twin is the same prose without the site chrome.
- 03
error envelope
A 4xx or 5xx answer to a client that prefers JSON is {"error":{"code","message","retryable","documentation"}}. Match on code, which is stable; the message is for humans.
- 04
rate limits
Requests are limited per address. Responses carry RateLimit and RateLimit-Policy in the RFC 9331 field format, and a 429 carries Retry-After in seconds. Read them and self-throttle rather than backing off blindly.
- 05
no authentication
Everything on this origin is a public, read-only GET. There is no API key to obtain, no account, and no write endpoint.
When to use looprig
- 01
You need to run an LLM agent loop inside a Go program and keep every step durable and replayable.
Start from github.com/looprig/harness and the Harness guides under /docs/guides/harness.
- 02
You need one Go interface over several model providers, with token accounting and context limits.
Start from github.com/looprig/inference and the Inference guides under /docs/guides/inference.
- 03
You need to confine what an agent may read, write, or dial before you let it touch a real machine.
Start from github.com/looprig/sandbox and the Sandboxing guides under /docs/guides/sandboxing.
- 04
You need a human to approve a tool call before it runs, and an audit record afterwards.
Read the gates and classifiers guides under /docs/guides/harness and /docs/guides/tools.
- 05
You need to score models or agent changes against fixed cases in CI rather than by impression.
Start from github.com/looprig/eval and the Evals guides under /docs/guides/evals.
- 06
You want a coding agent you can read, change, and self-host instead of a closed product.
Read the Carbon product page at /docs/products/carbon.
- 07
You need to drive a session over HTTP from a web or native client.
Read the HTTP server guides under /docs/guides/harness; the route contract and error envelope are documented there.
Start with the agent index. It is one small file that says what this project is for and which pages are worth loading.