BoonyardNN

Append-only shared memory substrate for multi-agent collaboration.

The primitive is an entry: a timestamped piece of text, written by a named agent, optionally threaded to a previous entry, optionally tagged. Entries are never destroyed and never silently changed — corrections happen by writing new entries that reference old ones. One table, a handful of well-chosen indexes, and the discipline to write into it. Your agents — human and AI alike — get a shared, permanent, queryable wall. That's the whole magic.

what it is

per-project
Each project owns one node — one SQLite file plus a small config. Your coding agent, your chat agent, and you all write into the same journal. The project's memory outlives every session that touched it.

in-project
Embed the package inside a project and run a node as the project's own append-only event log. Zero runtime dependencies — Python 3.11+ stdlib only — so it vendors in with a single cp -r.

over-many
An aggregator opens many nodes read-only and unions their entries. One view across every project's memory. It never writes; writes always belong to one node.

what it isn't

Not a personal AI agent — it's the journal your agents share, not an agent itself. Not a vector database — no embeddings by default, ever. Not a kitchen-sink platform — no schedulers, no gateways, no built-in LLM. Not telemetric — the package phones home to nothing. The substrate is small on purpose; agents bring their own everything else.

no lock-in, ever

The data is a SQLite file. Download it, open it in any SQLite tool, point your own server at it, copy it to another machine, fork the code, walk away. Exit cost is cp of one file. The eventual hosted service runs the same open-source package you can pip-install — convenience is the product, never the algorithm.

status

● pre-phase-1 — the substrate is designed; package extraction is next.
● the design canon (charter, 10 ADRs, 9 architecture docs) is complete.
● a live prototype has been running across the author's projects since 2025.
● open source (Apache 2.0) on release.