Projects building on IWE — a running list

[gimalay]

While sweeping GitHub recently I found more projects building on IWE than I knew existed. If you're building something on IWE, or I missed yours, reply and I'll add it.

Agent memory & AI integrations

• pi-iwe-kb by @ldarren — makes IWE the persistent-memory backend for the pi coding agent: seven kb_* tools, /kb-* slash commands, and lifecycle hooks that auto-recall context via iwe retrieve and auto-ingest facts back into a markdown vault. It even ships a memory-test mode that truncates chat history to prove the answers really come from the vault.
• agent-memory by @dzackgarza — goes deeper than anyone: binds the liwe Rust library in-process as the vault/graph/refactoring layer of a personal agent-memory system. The design transcript that led there is a great read.
• OKEngine by @jalewis — bakes pinned, checksum-verified iwe binaries into its Docker images as the agent-accessible graph layer of an LLM-maintained wiki engine (paired with qmd for hybrid search), documented in its supply-chain notes.

Knowledge tools

• comma by @kiil — a nushell editorial module using IWE as its notes/persistence layer, with a dedicated Why IWE page in its docs.
• percybrain by @percy-raskova — a Neovim Zettelkasten distribution with IWE as its LSP core.
• offshoreai by @mikekelly — a legal-knowledge agent project that adopted iwe for corpus stewardship: committed .iwe/config.toml, and a SETUP.md that teaches contributors the extract/inline/rename/stats workflow — deliberately developer-side, off the agent's runtime tool surface. A pattern worth studying.

Ships or bundles IWE

• coyote — bundles the iwe MCP server
• carbide — bundles iwes with its own default config and transform examples
• rusty-biscuit — its design docs build on and plan to extend iwes

Borrowed the ideas, credited the source

• MinionOS — ported three IWE graph algorithms (neighborhood expansion, reference-count ranking, squash) into its Python memory OS — see its analysis and changelog entry
• sverklo — adopted the squash/tree progressive-disclosure pattern and IWE's inclusion-vs-reference edge split, credited in code and changelog
• repoforge — credits IWE for squash, polyhierarchy, and depth-configurable retrieval in its competition analysis, with squash on its roadmap

A note on stability, since it's the honest question every embedder hits: the liwe API isn't stable yet — pin your versions. The CLI and MCP server are the practical integration surfaces today, and knowing what you depend on helps us not break it — that's half the reason for this thread.