brain
Personal agent infrastructure. A persistent memory layer that AI agents can depend on, with markdown exfiltration so the same content is also easy for humans to read and grep.
brain is a Go fork of beads. It keeps beads' versioned Dolt substrate and graph-shaped issue model, then adds:
- A unified surface for tasks, knowledge entries, and ISAs. One bag of docs discriminated by
kind (task | knowledge | both | isa), one set of verbs, one query layer.
- Markdown exfiltration. Every entry is rendered to a markdown file under
~/data/knowledge/entries/{kind}/{slug}.md on every write. Dolt is the source of truth; markdown is the human-readable view.
- ISA primitives. First-class support for PAI Algorithm v6.4+ ISAs (Ideal State Architecture documents) — including
bd new isa, bd isa-section, bd isa-render, and per-ISA section UPSERT semantics.
How it ships
One Go binary, two install names:
bd — full beads-compatible surface for project-local issue stores.brain — wrapper script that pins BEADS_DIR=$HOME/data/knowledge/.beads and dispatches the brain verbs at the top level. Argv[0] dispatch via BD_NAME=brain.
Verbs hoisted to the top level (no bd brain <verb> subtree):
bd new <kind> <title> [--slug=...] # create issue / knowledge / isa
bd link <from> <to> --type=<edge> # graph edge
bd related <id> # walk the graph
bd recast <id> --kind=<new-kind> # change kind discriminator
bd patch <id> --field=... --value=...
bd isa-section <id> <name> --value-stdin
bd isa-render <id>
See divergence/ for the full design history vs upstream beads.
Status
Active personal infrastructure. APIs may break without notice. No external contributions are accepted at this time — this repo exists to let agents I run depend on it, not as a community project.
Upstream
Forked from gastownhall/beads. Periodic rebases bring beads fixes forward. Brain-specific code lives under cmd/bd/brain_*.go, internal/brain/, and internal/storage/embeddeddolt/.
License
Inherits the upstream beads license. See LICENSE.
