tools · the open-source stack

Tools we build and share.

Everything we use in client work is built on our releases of free software. Patent-free. The knowledge of how to use them well, suited to your context, is what matters — the tools themselves are commons.

Twelve tools · grouped by what they do

Reason about plans, rules, and protocols

defeasible logic · multi-agent coordination · safe extension

agents talking in logical sentences

A shared, encrypted theory of facts, rules, and signed promises. Conclusions explain themselves; commitments become fulfilled or violated from the evidence.

multi-agent plan coordination

A plan your agents can reason about. Defeasible logic in SPL: readiness derives, blockers defeat, agents claim and merge through isolated worktrees. Twenty-one LLM providers.

defeasible reasoning engine

Reason about rules that have exceptions. Non-monotonic logic in Rust — conflict, priority, withdrawal, with a proof tree you can show to an auditor.

agent communication language

An agent communication language that extends itself, safely. Deterministic context-free, formally verified in Lean 4, accepted at IEEE S&P 2026.

routing fabric for agent traffic

Make any process an agent. cbcl-router dispatches asks to dialects on Erlang/OTP; hark gives every host one daemon and many concurrent agents — bash, Python, Claude, anything that connects to a socket.

Surface what matters

ambient retrieval · team knowledge · structured crawling

ambient relevance for ecosystems

Prior work that surfaces as you work. A UserPromptSubmit hook injects context into every prompt — trajectory-aware, mode-shaped, corrected by feedback.

team wiki for agents

A wiki whose answers show their work. Plain Markdown, MCP-native for agents, every claim sourced to the line it came from. Local-first.

crawler for the agent holding the leash

Clean structured output, no LLM in the tool, no vendor lock-in in your pipeline. The crawler decides nothing; the agent decides everything.

Pace, sense, and verify

somatic-first design · consent-scoped · zero-knowledge identity

AI that paces to the human

Five working modes, defeasible governance, transition friction. Pull-only, never interrupting. Your nervous system sets the tempo, not the chat window.

HRV as a research instrument

Heart-rate variability from a Polar H10 as a 3D phase-space trajectory. On-device, ESL-A licensed, used when the research scope warrants it.

zero-knowledge biometric auth

Prove who you are without revealing your biometrics. Halo2 ZK proofs over face templates, offline NFC/BLE verification, no central database.

near-zero-cost decentralised identity

Create, rotate, and update identity without transaction fees, a validator fleet, or a confirmation wait. Work offline; converge later.

Recent work

Taking shape in the open.

The latest changes from our research and engineering repositories.

Loading recent updates…

Explore the repositories ↗ Subscribe via Atom ↗

How the stack hangs together

Each tool stands alone. The pairings do extra work.

Common substrate

Defeasible logic shows up across elephant, hence, spindle, vibe harness, and zetl — rules that can yield, with proofs you can read. The same shape recurs because the shape is what's load-bearing.

Shared session boundaries

Sense, vibe harness, and ebs all key off a session ID. Mode signals reshape retrieval; biosignal trajectories align to mode transitions; nothing is forced together but everything is joinable.

Local-first by default

Every tool runs on your laptop. Cloud surfaces are optional infrastructure, never captive dependencies. Data lives on your disk in formats you can cat.

Patent-free, open licences

Apache 2.0 or AGPL-3.0 for most; ESL-A for the somatic and consent-scoped tools (ebs, vibe harness). The licence reflects what each tool is for.

Receipts where it counts

Hence shows you why a task is ready. Zetl cites the line behind every claim. Spindle prints the proof tree. Vibe harness traces every nudge. Sable verifies a proof, not a database.

Used in client work first

Tools are built because the engagement needed them. The release is what's left after the work — generalised, documented, and given back. Anuna's research practice keeps them honest.