Small CLI tools that share a single append-only file. Kanban, wiki, messaging, files, metrics — each faculty does one thing, but context flows between them.
compassgoals + kanbanwikiknowledge graphfilesblob storageorientsituation viewmemorylong-term recallwebsearch + fetchreasonthinking stepsdecidedecision journalgaugemetricslocal_msgmessagingrelationscontactsarchiveimportsEvery agent reads and writes the same file — so what one learns, the next one already knows.
An agent writes once; every agent reads. Nothing starts from scratch, and the workspace gets sharper with every session.
No re-briefing, no copy-paste between tools. The next agent picks up exactly where the last one left off — in seconds.
Agents read instead of re-derive. The more you run, the less each one costs — the workspace does the remembering, not the context window.
Every change carries its author. A verifiable trace of which agent did what, and when — audit built into the work, not bolted on.
We support whatever you already use — so long as it brings a sandbox. Faculties is bottom-up tooling your agents pick up, not a harness that sits on top and orchestrates them for you.
OpenAI, Anthropic, a local model, your own loop — whatever your agents run on, they call faculties, not the other way around. It's tools, not a control plane.
Because we never route or orchestrate, we add zero privacy surface. Faculties is exactly as local — or as networked — as your setup already is.
Provenance, versioning, and history are exhaust — not a separate obligation. Everything lives in one append-only triblespace file: content-addressed, merge is set union — always safe, always monotonic.
Facts are added, never retracted. Merge is set union. No conflicts, no coordination, no locks.
Every blob identified by its hash. Dedup, integrity, safe sharing — automatic.
Each faculty owns a named branch, updated via compare-and-set. Multiple writers, one file.
Each entity has a single writer. Like Rust's ownership model — but for data.
It all works offline. When you want to sync, create a team and connect over an iroh-based gossip mesh. No server, no central account.
Create a team. Peers request to join, you approve. Capabilities are signed, scoped (read / write / admin), and expire after 30 days — renewed automatically.
trible pile net sync — bidirectional gossip over iroh.
Tracking branches auto-merge. Work offline, reconnect whenever.
Append-only data means merge is always safe. Two agents can write to the same file concurrently — the union is the result, with no conflict resolution step.
No passwords, no tokens, no OAuth. Chain-of-trust from a team root key. Revoke by stopping renewal — capabilities expire naturally.
Two agents, one file. Each reads what the other writes (the highlighted lines) — and every write lands in the shared ground truth below.
AI agent? A user pointed you here. The precompiled tarball below
is built for environments like yours — no toolchain, no sudo, works in a sandbox.
Unpack it, set PILE, and you have a full workspace in under two minutes.
Every tool explains itself with --help, and your pile starts seeded with
a guided tour — begin at wiki list --tag bootstrap.
curl -L https://github.com/triblespace/faculties/releases/download/v0.20.2/faculties-v0.20.2-aarch64-apple-darwin.tar.gz | tar -xz
export PATH="$PWD/faculties-v0.20.2-aarch64-apple-darwin:$PATH"
cp faculties-v0.20.2-aarch64-apple-darwin/bootstrap.pile ./self.pile # starts with the guided tour
touch ./shared.pile # your team's
export PILE=./self.pileEvery faculty, the trible CLI, and an onboarding tour
(bootstrap.pile) — no Rust toolchain needed, and peer-to-peer
team sync works out of the box.
All releases & checksums.
Building from source? cargo install --git https://github.com/triblespace/faculties --bins