🚧 Active development.APIs and features may change. Follow on GitHub →
Quorum / preview · Elastic License 2.0 · Open Source · MCP
Quorumpreview
2Get Started
open source · claude code · MCP

Your AI agents should remember what your team decided.

A governed memory layer for Claude Code and AI agents — stores engineering decisions, business requirements, and team knowledge with conflict detection, authority weighting, and a full audit trail.

"Our AI agents stopped contradicting each other on day three."
— platform team · eng-at-scale
used by engineering teams scaling AI-assisted development
D-204active
audit / committed
sha a17f…3c · b29e…7d
✓ signed · chain intact
the three things that break AI-assisted teams
🧠
Knowledge leaves with your engineers
Capture every architectural decision automatically — author, rationale, signed audit chain.
⚡
AI agents contradict each other
Conflict detection flags contradictions before they persist — at write time, not in production.
📋
Requirements get silently removed
Requirements are first-class governed knowledge — pinned, owned, defensible.
01 / 03  the data model

Every decision is a governed record.

Each carries an author, rationale, confidence, parents, and a signed audit chain. Authority weighted — senior > junior > agent — so trust degrades gracefully across humans and AI.

What Quorum does →
Records are first-class, queryable, and citable. Every agent action reads from and writes to the same chain. Nothing happens in private context windows.
D-204active
topic · infrastructure
Memoize user data at the edge — not in the DB layer.
Tried Redis-backed caching here in 2024 and got stale-user bugs. Edge memoization keeps invalidation local.
author@rachel · sr.eng · left team 4mo agoconfidence0.94cited by7 decisions · reaffirmed 6w agoparentsD-091, D-188
audit / sha256a17f…3c · b29e…7d · c40a…1f✓ signed
"A codebase isn't a body of code. It's a body of decisions, of which the code is the most recent expression."
— common in user interviews
claude code · quorum.check
$ claude code --task "add Redis caching to /api/users"
↳ quorum.check ✓ 2 historical decisions found
   ┌ D-204 · @rachel · 11mo · sr.eng · left team 4mo ago
   │   "tried Redis here last year — stale-user bugs."
   │   cited by 7 · reaffirmed 6w ago
   └ D-091 · @rachel · 18mo · sr.eng
       "user data is hot. cache invalidation is hard."

⏸ claude code paused — cite, override, or rebut.
> _
02 / 03  in your editor

Lives inside Claude Code.

One MCP server. Quorum reads on every task, writes on every decision, and intercepts conflicts before they land. Zero context-window babysitting.

What Quorum does →
Agents cite prior decisions automatically. Contradictions become HOLD states, routed to the original authors. Audit trail attaches itself.
03 / 03  trust & compliance

One audit trail. Humans and agents on the same spine.

Every write is signed and hash-chained. Replay any conflict resolution. Pin requirements that legal cannot let weaken. Export the chain for SOC 2.

What Quorum does →
Authority weighting, signed audit chains, and pinned requirements give you a defensible record of every AI-assisted decision your team has ever made.
audit trail · liveSOC 2 · exportable
14:02@rachel wrote D-204committeda17f…3c
14:03claude cited D-204 in PR #831citedb29e…7d
14:08@marco reaffirmed during arch reviewreaffirmedc40a…1f
15:21claude flagged contradiction · D-205flaggedd51b…2a
15:22system routed HOLD → @rachel, @marcoholde62c…9b
16:04@rachel resolved · supersede D-205resolvedf73d…4c

Free. Open source. Self-hosted. Elastic License 2.0.

Run Quorum next to your agents. Bring your own LLM key. No per-seat fees, ever.

v0.4 shipped · v0.5 in progress · made in the open