Open Source

Our Open Source Philosophy

We build on open source and we give back to open source. This isn’t a marketing strategy - it’s how good software gets built.

What We Open Source

SDKs and Client Libraries

All our client libraries are open source (MIT licensed):

• Python SDK
• TypeScript/Node SDK
• Go SDK (coming)

Customers should be able to read, audit, and contribute to the code that runs in their infrastructure.

Evaluation Frameworks

Our core evaluation primitives - metrics, scorers, and test runners - are open source. This includes:

• Standard evaluation metrics (factuality, relevance, toxicity, etc.)
• Dataset utilities for managing test data
• CLI tools for running evaluations locally

Integrations

Integrations with popular AI frameworks (LangChain, LlamaIndex, CrewAI, AutoGen) are open source and community-maintained.

Research and Datasets

When we publish research, we release the datasets and code to reproduce our results. No “available upon request” - it’s on GitHub.

What We Keep Proprietary

• Turing evaluation models - our in-house eval models, fine-tuned on proprietary data
• Falcon AI - the in-platform AI assistant that reads your evals, traces, and datasets to answer questions and run actions
• Optimization - RL-based prompt and model optimization trained on platform usage data
• Simulation - synthetic user generation, scenario libraries, and agent simulation runtime
• AI-powered annotations - auto-labelling and trace annotation powered by our internal models
• Prompt assistance - prompt rewriting, diffing, and suggestion features inside the Agent IDE
• Agentic evals - multi-step, tool-using evaluators that reason over full agent trajectories
• Enterprise features - SSO, RBAC, audit logs

This is the open-core model: the tools and primitives are free and open, the models and AI-powered features are a paid product. The line we hold is simple - if it’s something you’d run in your own code, it’s open source. If it’s something we trained, hosted, or built on top of customer signal, it’s part of the paid product.

Contributing to Other Projects

We actively contribute to the projects we depend on:

• Bug fixes, performance improvements, and documentation
• Financial sponsorship of critical dependencies
• Upstream first - we contribute fixes to the source rather than maintaining private forks

How to Contribute to Future AGI

Issues

• Check existing issues before filing a new one
• Include reproduction steps, expected behavior, and actual behavior
• Label requests with bug, feature, or docs

Pull Requests

• Fork the repo, create a branch, make your changes
• Write tests for new functionality
• Follow the existing code style (we have linting for this)
• Keep PRs focused - one change per PR
• Write a clear description of what and why

Community Standards

• Be respectful and constructive
• No spam, self-promotion, or off-topic content
• We follow the Contributor Covenant code of conduct

Community Channels

Channel	Purpose
GitHub	Code, issues, PRs, discussions
Discord	Real-time chat, help, community
Blog	Long-form content, tutorials, updates
Twitter/X	Announcements, hot takes