No tool in this category was invented from nothing, and a memory project of all things should remember where its ideas came from. This page credits the people, agents, research, and peers Bourdon is built on. If we learned something from you and you’re not here, that’s a bug; write us at [email protected].
Maintainer
Ryan Davis, RADLAB LLC. Project lead, thesis author, sets direction, merges every PR.
AI co-implementors
Bourdon is co-built with AI agents working alongside their humans, under an agent-as-author convention: commits land under a human identity, PRs name the agent and link the originating session. Several agents wrote their own Bourdon integrations: the tools that built it are the tools it gives a shared memory to.
- Claude Code (Anthropic): core orchestrator, L0/L1 logic, federation (L5/L6), spec, tests.
- Codex (OpenAI): co-implementor on the reference orchestrator and CLI, and co-author of its own participant; hybrid memory-cycle tooling, MCP smoke assertions, CI automation.
- Cursor Cloud Agent: the Cursor participant, including the SQLite session reverse-engineering.
- GitHub Copilot: authored its own participant and CLI subcommands, designing its convention-file memory layer from the inside out.
- Cascade (Windsurf): authored its own participant plus the unified
bourdon doctorandbourdon export-allcommands. - Hermes: RADLAB’s in-house Slack agent, which wired up its own connection to the library.
Research this work stands on
Bourdon’s vocabulary deliberately maps to the wider agent-memory field; the full cross-reference lives in RELATED_WORK.md. The short version of the debt:
- Memora: abstraction and cue-anchor framing that informs our L0/L1 split.
- Letta: editable memory blocks, the closest analogue to our entity synopses.
- Zep / Graphiti: temporal validity windows, which our
valid_from/valid_tofields adopt directly. - Mem0: the 3-tier scope model and the compliance bar for the category.
- Intrinsic Memory Agents: agent profiles, the ancestor of our L5 manifest.
- Memory in the Age of AI Agents: the survey whose taxonomy we cross-reference against.
- Shared Context Store proposal: the formal MCP-era framing closest to our L6 federation library.
Peers sharpening the category
Other teams are making the case that AI memory should belong to the user, each with a different architecture: Second Brain for AI, who shipped the self-custody argument before we launched, Glia, and Atlaso. Competition here is validation; our honest side-by-sides live at /compare.
Built with
Python and TypeScript. MCP for the agent surface. The tray app is Tauri; this site is Next.js on Cloudflare Pages. The name is the deep continuous drone of a pipe organ: the foundational tone that holds every other voice in place.