about
An agent harness you can own.
looprig is an agent harness built as independent, stdlib-first Go modules under the Apache License 2.0, for teams that need to own the runtime their agents run on.
What looprig is
What looprig is
looprig is a set of independent Go modules for building and running LLM agents. The harness owns the loop: it turns a model, a set of tools, and a workspace into a session whose every step is journaled, resumable, and inspectable. Around that core sit modules for model access, credentials, secrets, storage, sandboxing, evaluation, and the protocol adapters an agent needs to talk to editors and other agents.
Each module is its own repository with its own tests, its own release tags, and a public boundary that is documented page by page. You can adopt one module without adopting the rest: the inference gateway works without the harness, the sandbox works without either, and the storage interfaces are implemented by separate backend modules rather than baked into the runtime.
How it is built
How it is built
The modules are stdlib-first Go. Dependencies are pinned by version and verified by checksum rather than vendored, so a build is reproducible and a dependency upgrade is a visible, reviewable change. Releases move from the leaves of the dependency graph upward, and a dependent may only name a version that already exists upstream.
The documentation is held to the same standard as the code. Technical claims on this site are derived from implementation code, tests, module files, and immutable release tags, not from plans or roadmaps. Every documentation page is published with a raw Markdown twin so that a machine reader gets the same text a human reads, without the surrounding markup.
Licence and source
Licence and source
Every published module is released under the Apache License 2.0 at github.com/looprig. The source, the tests that back the documentation, and the release tags are all public. There is no hosted looprig service and no account to create: you run the software yourself, on your own machines, against your own model providers.
The project is maintained from the United States. Questions, bug reports, and security disclosures are welcome at crew@looprig.com.