# ICRL

## Docs

- [Agent](https://icrl.dev/docs/api-reference/agent.md): Main Python agent class
- [AnthropicVertexProvider](https://icrl.dev/docs/api-reference/anthropic-vertex-provider.md): Python provider for Claude models on Google Vertex AI
- [CurationManager](https://icrl.dev/docs/api-reference/curation-manager.md): Automatic pruning manager
- [Embedders](https://icrl.dev/docs/api-reference/embedder.md): Python embedding implementations and defaults
- [Environment Protocol](https://icrl.dev/docs/api-reference/environment.md): Required environment interface
- [LiteLLMProvider](https://icrl.dev/docs/api-reference/litellm-provider.md): Built-in Python provider powered by litellm
- [LLMProvider Protocol](https://icrl.dev/docs/api-reference/llm-provider.md): Required Python provider interface
- [Message](https://icrl.dev/docs/api-reference/message.md): LLM message model
- [API Overview](https://icrl.dev/docs/api-reference/overview.md): Python and TypeScript API surfaces
- [Step](https://icrl.dev/docs/api-reference/step.md): Single agent step
- [StepContext](https://icrl.dev/docs/api-reference/step-context.md): Prompt-formatting context for one step
- [Trajectory](https://icrl.dev/docs/api-reference/trajectory.md): Episode record model
- [TrajectoryDatabase](https://icrl.dev/docs/api-reference/trajectory-database.md): Python trajectory store, embedding index, and curation metadata
- [TrajectoryRetriever](https://icrl.dev/docs/api-reference/trajectory-retriever.md): Step-example retrieval helper
- [Agent](https://icrl.dev/docs/api-reference/typescript-agent.md): Main TypeScript agent class
- [Anthropic Provider](https://icrl.dev/docs/api-reference/typescript-anthropic-provider.md): Anthropic and Anthropic Vertex providers for TypeScript
- [CurationManager](https://icrl.dev/docs/api-reference/typescript-curation-manager.md): Automatic curation for trajectory databases
- [Embedder Protocol](https://icrl.dev/docs/api-reference/typescript-embedder.md): Required embedding interface for TypeScript
- [Environment Protocol](https://icrl.dev/docs/api-reference/typescript-environment.md): Required environment interface for TypeScript
- [LLMProvider Protocol](https://icrl.dev/docs/api-reference/typescript-llm-provider.md): Required TypeScript provider interface
- [Models](https://icrl.dev/docs/api-reference/typescript-models.md): Trajectory, Step, Message, and related types
- [OpenAI Provider](https://icrl.dev/docs/api-reference/typescript-openai-provider.md): OpenAI LLM and embedding provider for TypeScript
- [TypeScript Package API](https://icrl.dev/docs/api-reference/typescript-package.md): Overview of the ICRL TypeScript library
- [ReActLoop](https://icrl.dev/docs/api-reference/typescript-react-loop.md): ReAct-style agent loop with planning, reasoning, and acting
- [StorageAdapter](https://icrl.dev/docs/api-reference/typescript-storage.md): Pluggable storage interface for trajectories and embeddings
- [TrajectoryDatabase](https://icrl.dev/docs/api-reference/typescript-trajectory-database.md): TypeScript trajectory store with pluggable storage and semantic search
- [TrajectoryRetriever](https://icrl.dev/docs/api-reference/typescript-trajectory-retriever.md): Step-level retriever for in-context learning
- [Contributing](https://icrl.dev/docs/contributing.md): Project structure, setup, and development workflow
- [Curation](https://icrl.dev/docs/core-concepts/curation.md): How ICRL prunes low-utility trajectories
- [ICRL Algorithm](https://icrl.dev/docs/core-concepts/icrl-algorithm.md): How ICRL bootstraps agent performance with self-generated trajectories
- [ReAct Loop](https://icrl.dev/docs/core-concepts/react-loop.md): Execution model used by ICRL agents
- [Trajectory Database](https://icrl.dev/docs/core-concepts/trajectory-database.md): How trajectories, embeddings, and curation metadata are stored
- [Basic Anthropic Demo](https://icrl.dev/docs/examples/basic-anthropic-demo.md): Minimal ICRL run with Anthropic — verify your Claude setup
- [Basic OpenAI Demo](https://icrl.dev/docs/examples/basic-openai-demo.md): Minimal ICRL run with OpenAI — the fastest way to verify your setup
- [Codebase Patterns Demo](https://icrl.dev/docs/examples/codebase-patterns-demo.md): ICRL learns your team's APIResponse, exception handling, and service layer patterns
- [Exception Handling Demo](https://icrl.dev/docs/examples/exception-handling-demo.md): ICRL vs vanilla: applying past precedents when policy is too rigid
- [File System Agent](https://icrl.dev/docs/examples/file-system-agent.md): Python example using a simulated filesystem environment for training and evaluation
- [Harbor Coding Agent](https://icrl.dev/docs/examples/harbor-coding-agent.md): Python coding-agent example aligned with Harbor-style sandbox workflows
- [IT Support Demo](https://icrl.dev/docs/examples/it-support-demo.md): ICRL vs vanilla: learning from past support tickets when docs are incomplete
- [Preference Learning Demo](https://icrl.dev/docs/examples/preference-learning-demo.md): ICRL adapts to user styles — terse expert vs detailed learner vs executive summary
- [Testing With Mock LLM](https://icrl.dev/docs/examples/testing-with-mock-llm.md): Develop and test ICRL flows without external model calls
- [TypeScript Demos](https://icrl.dev/docs/examples/typescript-demos.md): End-to-end Agent examples with OpenAI, Anthropic, and Convex storage
- [Batch Training](https://icrl.dev/docs/guides/batch-training.md): Train and evaluate across many goals
- [CLI](https://icrl.dev/docs/guides/cli.md): An interactive coding agent in your terminal — like Claude Code, but it learns from every successful run
- [Convex Provider](https://icrl.dev/docs/guides/convex-provider.md): Using the ConvexAdapter storage backend with the TypeScript package
- [Custom Environments](https://icrl.dev/docs/guides/custom-environments.md): Implement the Environment interface for your domain
- [Custom LLM Providers](https://icrl.dev/docs/guides/custom-llm-providers.md): Implement your own provider for Python or TypeScript
- [Prompt Templates](https://icrl.dev/docs/guides/prompt-templates.md): Design prompts for planning, reasoning, and action generation
- [TypeScript Package](https://icrl.dev/docs/guides/typescript-package.md): How to use the icrl TypeScript library
- [Web Example Setup](https://icrl.dev/docs/guides/web-example.md): Detailed setup for the Next.js + Convex web demo
- [Installation](https://icrl.dev/docs/installation.md): Install and configure ICRL for Python, TypeScript, CLI, and the web demo
- [Introduction](https://icrl.dev/docs/introduction.md): ICRL for Python and TypeScript: self-improving agents that learn from successful trajectories
- [Quickstart](https://icrl.dev/docs/quickstart.md): Run your first ICRL agent in Python or TypeScript

## OpenAPI Specs

- [openapi](https://icrl.dev/docs/api-reference/openapi.json)

## Optional

- [GitHub](https://github.com/SuperAce100/icrl)
- [PyPI](https://pypi.org/project/icrl-py/)
- [npm](https://www.npmjs.com/package/icrl)