Pi is a TypeScript-based AI agent toolkit that combines a unified LLM API, agent runtime, terminal UI, and an interactive coding agent CLI.
Pi is an agent harness project centered on a self-extensible coding agent. The repository includes separate packages for a unified multi-provider LLM API, an agent runtime with tool calling and state management, an interactive coding agent CLI, and a terminal UI library. The README also lists related packages for durable workflows, telemetry, and application composition.
It addresses the need to build AI agents and coding assistants on top of multiple model providers without tying an application to a single LLM vendor. It also aims to provide the surrounding agent infrastructure needed for interactive coding, terminal-based workflows, and stateful tool use. The README indicates that it is meant for users who may need stronger isolation as well, since it does not provide built-in permission boundaries for filesystem, process, network, or credential access.
At a high level, Pi is organized as a set of packages that each cover a layer of the agent stack: model access, agent runtime, terminal interaction, and coding-oriented CLI workflows. The README says the agent runtime handles tool calling and state management, while the AI package abstracts multiple providers such as OpenAI, Anthropic, and Google. For safer deployment, the project points users toward containerization or sandboxing patterns when they need stronger security boundaries.
Its attention appears to come from being a broad agent toolkit rather than a single-purpose demo: it combines model abstraction, agent runtime, a TUI, and a coding agent CLI in one project. The repository also emphasizes practical concerns such as release builds, offline build paths, supply-chain hardening, and containerization guidance, which can make it appealing to developers looking for a production-minded starting point. The README explicitly calls out public OSS session sharing for improving real-world coding agents, which suggests an active effort around community usage and data collection.
The README itself points to related projects in the same ecosystem, including pi-chat for Slack/chat automation and workflows, as well as sibling packages for telemetry, durable runtime, terminal UI, and application composition. More generally, the project sits in the space of AI agent frameworks and coding-agent CLIs, but the README does not name direct competitors. Based on the text alone, the clearest comparisons are other agent toolkits and model-agnostic LLM wrappers.
AI-explained · grounded in each repo's README