Skip to documentation
Documentation navigation

Documentation navigation

developer documentation

Build with Looprig.

Find the shortest path from an empty workspace to a durable, confined, gated, transparent agent. Browse by task, module, or package reference.

working as an agent? start at /llms.txt; every page also has a raw Markdown twin at its stable URL plus .md.

Looprig is a collection of composable Go modules for building model-powered applications. Start with a provider-neutral model call, add Harness when you need an agent runtime, then select tools, storage, sandboxing, workflows, protocols, and interfaces according to the product you are building.

Build with Looprig

GoalBegin with
Call OpenAI, Anthropic, Ollama, or another model through one contractInference
Run stateful turns, tools, events, sessions, gates, and delegationHarness
Add file, search, process, interaction, and task capabilitiesTools
Confine child processes and verify native guaranteesSandboxing
Add durable graphs, interruption, recovery, and workflow toolsWorkflows
Connect terminal or browser consumersTUI and Web UI

Start with an outcome

The Getting Started tutorial builds a coding assistant from an empty Go module. It connects a model, creates a Loop and Rig, runs a Session, adds read-only tools, persists history, manages a workspace, confines process tools, and exposes a CLI.

Every implementation step links to a runnable checkpoint and the deeper guide for its module.

Choose only what you need

The modules have narrow responsibilities and compose through explicit interfaces. An application can use Inference without Harness, Harness without a browser client, or the Web UI with any frontend framework. Add a module when the application needs the boundary it owns.

When you need API-level detail, use the module guides in the left navigation and follow their source and runnable-example links.