PARTNERS CIRCLE OPEN SOURCE

The clearest look at how we build

Tools we needed while building systems and could not buy off the shelf. The source is all on GitHub, free for anyone to use — and the closest look you can get at the standard of our work.

PROJECTS

MeMesh

Memory your AI agent doesn't lose between sessions

Every new conversation with an AI coding agent starts from zero — the architecture decisions, the bugs already fixed, the constraints already learned, all re-explained from scratch. MeMesh is a memory layer that runs on your own machine: one SQLite file, no cloud, no API key required, that captures this automatically and surfaces what's relevant before the agent starts working.

WHO IT FITS
Developers using Claude Code or any MCP-compatible tool
INSTALL
npm install -g @pcircle/memesh
  • 95.60% R@5 on the public LongMemEval-S benchmark (reproducible)
  • Local-first: one SQLite file, no cloud, no API key by default
  • The same memory is shared across Claude Code, Codex CLI, and Gemini CLI
  • Native integration with Hermes Agent (NousResearch) and OpenClaw: each has a first-party memory plugin at the same tier as their own built-in backends
Full details →View source on GitHub

Toonify MCP

Trims large tool output before it fills up the conversation

Commands and file reads in Claude Code often return large, messy output — full API responses, long logs, stack traces. Toonify MCP compresses this automatically before it enters the context window, running in the background with no change to how you already work.

WHO IT FITS
Developers running Claude Code against large JSON, YAML, or test output
INSTALL
git clone https://github.com/PCIRCLE-AI/toonify-mcp.git
  • Automatically compresses large JSON, YAML, and API responses
  • Trims long test failures and stack traces
  • Runs in the background — no change to your existing workflow
Full details →View source on GitHub

XylonStudio

Chip timing analysis: runs OpenROAD for real, reads back measured numbers, and repairs only with your approval

Give it RTL and timing constraints (SDC). XylonStudio runs a resource-bounded OpenROAD flow on your own machine and reads back native measurements — how late the worst path is (WNS), how late all violations are combined (TNS), and the worst path itself. When there is a violation it proposes one bounded repair based on that measurement, waits for your approval, reruns, and shows the before-and-after numbers side by side. Your design files never leave your machine.

WHO IT FITS
Engineers doing digital chip timing analysis, and students learning that work
INSTALL
git clone https://github.com/kevintseng/xylon-studio.git
  • Preflight before anything runs — if the environment, resources, or inputs are wrong it stops there, not halfway through
  • One repair at a time, applied only when you approve it; then a rerun and a side-by-side comparison
  • A number it could not measure is never backfilled from another one, and a run that failed is never shown as done
  • Its own stated boundary: timing analysis and bounded repair — no claim of automatic timing closure or foundry signoff
Full details →View source on GitHub

WHY WE OPEN-SOURCE

Problems we solved for ourselves, shared so others skip the same detour

Each of these started as a real problem in front of us, not a planned open-source release. The code is public — use it directly, read it, or open an issue. This is not a paid service: no customization quotes, no support contracts. If you need us to build a system for you, see our services.

The source is on GitHub

Free to use, source fully public. Reach out if you want to know how it's built, or about working together.

GitHub →About Partners Circle