TypeScript agent toolkit with supply-chain paranoia

What it does

Pi is a TypeScript monorepo that combines a multi-provider LLM client (pi-ai), an agent runtime with tool calling (pi-agent-core), and an interactive coding-agent CLI (pi-coding-agent). It also ships a differential-rendering terminal UI library (pi-tui). The project doubles as a research platform: maintainers ask users to publish real-world OSS session logs to Hugging Face so the agent can learn from actual failures instead of polished benchmarks.

The interesting bit

Most agent repos bury dependency management in a footnote; Pi dedicates an entire section to supply-chain hardening. Direct external deps are pinned to exact versions, lockfile changes are blocked by pre-commit hooks unless explicitly overridden, and releases are smoke-tested in isolated npm and Bun installs before they ever get tagged.

Key highlights

• Unified LLM API supporting OpenAI, Anthropic, Google, and others via pi-ai
• Agent runtime with tool calling and state management in pi-agent-core
• Interactive coding-agent CLI with differential-rendered TUI
• Published CLI ships with an npm-shrinkwrap.json to pin transitive dependencies
• Auto-closes issues and PRs from new contributors by default; maintainers review daily

Caveats

• New contributors face an auto-closed door by default, which the README mentions without apology.
• Tests skip LLM-dependent suites if you lack API keys for the providers it wraps.

Verdict

Worth a look if you want a hackable, TypeScript-native agent stack and appreciate maintainers who treat dependency auditing as a lifestyle. Skip it if you need an open-door contribution policy or are allergic to supplying your own LLM API keys.