pi
Named by 44 kept items in this workspace.
32 items name it under frameworks and tools, 8 under classification.technologies, 3 under classification.key entities and 1 under models.
Search every workspace for pi. It matches the name as written, so it can find other things that share it.
The items that name it 44
Runs the core agentic loop of the Minecraft harness.
Agent harness LFM3-1.7B is post-trained to run in.
Agentic harness LFM3-1.7B is post-trained to run in.
Agent harness used with Opus 5.5 in the MemoryTax test.
Coding agent that Sphinx works with.
Buterin's agent harness; the privacy wrapper ("paranoid mode") is built for it and it runs inside a bubblewrap sandbox
Agent harness used for the author's runs of OpenAI models on DeepSWE Python tasks
Agent harness used in agentic RL training and for local agents
A local coding agent that alook can connect to, giving it an identity/inbox so others can talk to it directly.
A "minimal agent harness" that other projects, such as Flue, have built on top of.
Minimal, extensible agent harness; Pi 1.0 adds native MCP support by default, Codemode, non-LLM/image model support, virtual-model extensions, deferred tool loading, cache warming, mid-conversation system messages
A minimal, elegant AI coding agent harness that the author customizes and tinkers with, avoiding lock-in to AI companies
Coding agent harness, discussed as structurally similar to Claude Code and Codex.
Agent that can be connected to Magnitude with one click.
Provider-scoped coding agent managed by magpie
The coding agent (MIT licensed) that mu is built on; its own commands (/model, /thinking, /login, /resume, /tree, /fork, /compact, /export) remain unchanged in mu.
The reference implementation coding agent that PiG follows; PiG's pinned reference release is 0.87.1 at a stated commit.
Coding agent supported via pi install npm:jev-guard, with block and confirm dialog.
The coding agent that pi-jev extends; runs tool calls (bash, write, edit) that pi-jev gates and judges.
coding agent that funes can be added to for memory/recall; can be used with a local model or one served through the Hugging Face router
Small extensible terminal coding agent working with Ollama, LM Studio, vLLM and other local servers
A minimal agent harness for AI-powered coding that runs in the terminal, supporting custom providers via a models.json config to connect to a local vLLM-served Granite 4.2 model.
Host agent harness; installs agy-staff with pi install and uses /skill:agy-<persona> skills.
Local agent engine option for BYOA; requires an explicit unsandboxed compatibility opt-in.
One of the AI coding agents antislop installs into and works with; the Pi package updates through Pi itself.
Coding agent that uses a single-file native extension (rather than the proxy) for seamless vision integration.
Additional supported host; same adapter also reads Oh My Pi (OMP) sessions
Coding agent supporting Superpowers as a Pi package; loads Superpowers skills plus a small extension injecting the using-superpowers bootstrap at session startup and after compaction; has native skills so no compatibility Skill tool is required.
AI coding agent tool integrated by CodeBurn.
A minimalist, self-modifying AI coding agent created by Mario Zechner; built with as few features as possible to behave in a stable, consistent way, and can modify itself so a user can create a bespoke harness for any task.
Skill-capable host supporting Ponytail commands
Coding agent configured to use the model via a local llama.cpp OpenAI-compatible server, installed via npm as @earendil-works/pi-coding-agent
All technologies in this workspace.