See It In Action
What's New in v1.9.0
- Web Dashboard (
/brain-dashboard) — browse, edit, and curate all agent memories at localhost:7432. Overview stats, memory table with inline editing, decision browser, and file map.
- Standup Meeting UI (
/brain-standup) — cinematic agent standup visualization at localhost:7433. Walk through each agent's completed work, blockers, and decisions with keyboard navigation.
- Approval Workflow (
/brain-approve) — memories from /brain-remember now stage as PENDING. Approve, reject, or flag via dashboard or CLI.
- Memory Quality Scoring — confidence levels (HIGH/MEDIUM/LOW/PENDING). Frequently accessed memories auto-promote; stale memories auto-demote.
- Better Knowledge Injection — KB chunks from agent
index calls injected into future teammates. Cross-team task visibility. Context budget doubled to 6000 chars.
- Smarter Agent Extraction — custom agent names and task subjects extracted from conversation transcripts.
/brain-learn now extracts 15+ conventions from git history.
Why claude-teams-brain?
Cut token usage by 90–97%
A single npm test dumps 20,000 tokens of passing tests into context. git push adds transfer stats nobody reads. Every wasted token costs money and shrinks your context window.
claude-teams-brain replaces raw Bash with token-efficient MCP tools that filter, index, and search command output — so only what matters enters context.
Persistent memory across sessions
Agent Teams are powerful — but ephemeral. Your backend agent spent two hours learning your conventions and building auth. Tomorrow, a new backend agent starts from zero.
claude-teams-brain indexes everything — tasks, decisions, files — per role. When backend spawns again, it receives the full history of what past backend agents built.
Smarter agents, better results
Agents that start with context make better decisions. Convention learning from git history, shared knowledge base between teammates, and role-specific injection mean your agents produce higher-quality output from the first command.
Install
One command:
npx claude-teams-brain
Or with curl:
bash <(curl -fsSL https://raw.githubusercontent.com/Gr122lyBr/claude-teams-brain/master/claude-teams-brain/scripts/install.sh)
Then restart Claude Code. That's it — the plugin activates automatically.
What It Does
Token-Efficient Command Execution
Every command through the brain's MCP tools passes through an 8-stage filtering pipeline with specialized parsers for 60+ commands:
| Command | Without brain | With brain | Savings |
|---|
npm test | 20,000 tokens of passing tests | Summary + failures only | 90%+ |
git push | Transfer stats, compression, deltas | ok main | 98% |
npm install | Warnings, progress bars, funding | added 542 packages in 12s | 90%+ |
pytest (all pass) | Full session output | 15 passed in 2.34s | 82% |
docker build | Layer-by-layer progress | Final image + errors | 95% |
tsc (no errors) | Verbose compilation | ✓ no errors | 90%+ |
Auto-indexing — every command's output is indexed into a searchable knowledge base. Run a command once, search it many times. Teammates share the same KB — no duplicate work.
5 MCP tools replace raw Bash for any command that produces output:
batch_execute — run multiple commands in one call, all output auto-indexedexecute — single command with optional auto-indexing for large outputsearch — query indexed output without re-running commandsindex — save findings for yourself and teammatesstats — see your token savings
Cross-Session Memory
Session 1 Session 2
───────── ─────────
You: "Build payments module" You: "Add refund support"
backend agent spawns (blank) backend agent spawns
↓ ↓
builds Stripe integration 🧠 Brain injects memory:
creates controller.ts • Past work: Stripe integration
decides: use PaymentIntents API • Files: controller.ts, stripe.service.ts
↓ • Decision: use PaymentIntents API
🧠 Brain indexes everything • Rule: all endpoints need auth
↓
picks up exactly where it left off
The brain hooks into 8 lifecycle events to capture and inject context automatically:
- SubagentStart — injects role-specific memory into new teammates
- SubagentStop — parses transcripts, extracts decisions, tracks files
- SessionStart — warms up KB with project context (CLAUDE.md, git log, directory tree)
- SessionEnd — compresses session into summary for future reference
- TaskCompleted — indexes individual task results immediately
Memory is role-based — backend, frontend, tests, devops each build their own history. When a teammate spawns, it receives only what's relevant to its role.
Works Without Agent Teams
You don't need Agent Teams to benefit. In solo mode:
- All 5 MCP tools work — same 90%+ token savings on every command
- Memory builds from your own sessions via TaskCompleted and SessionEnd hooks
/brain-learn, /brain-remember, and all commands work normally- Convention profiles and git history learning work the same way
Solo mode activates automatically when Agent Teams isn't enabled. No configuration needed.
Quick Start
Existing repo:
/brain-learn
Scans your git history and auto-extracts conventions, architecture, file coupling, and hotspots. Zero config.
New project:
/brain-seed nextjs-prisma
Loads pre-built conventions. Profiles: nextjs-prisma, fastapi, go-microservices, react-native, python-general.
Then just use Claude Code normally. Token savings and memory building happen automatically.
Commands
| Command | Description |
|---|
/brain-dashboard | Open web dashboard for reviewing and curating memories |
/brain-standup | Open cinematic standup meeting visualization |
/brain-approve | Approve, reject, or flag pending memories |
/brain-learn | Auto-learn conventions from git history |
/brain-remember <text> | Store a rule (staged as PENDING until approved) |
/brain-forget <text> | Remove a stored memory |
/brain-search <query> | Search the brain knowledge base |
/brain-query <role> | Preview what context a teammate would receive |
/brain-export | Export knowledge as CONVENTIONS.md |
/brain-stats | Full stats: memory + KB + token savings |
/brain-runs | List past sessions |
/brain-replay [run-id] | Replay a past session as narrative |
/brain-update | Pull latest version |
Key Features
| |
|---|
| Web Dashboard | Browse, edit, and curate all memories in a dark-themed web UI |
| Standup Meeting UI | Cinematic per-agent briefing with keyboard navigation |
| Memory Quality | Confidence scoring with auto-promote/demote lifecycle |
| Approval Workflow | Stage, approve, reject, or flag memories before injection |
| Token-efficient execution | 60+ command-aware filters, 8-stage pipeline — 90–97% token reduction |
| Auto-indexing KB | Every command output indexed and searchable. Teammates share results |
| Cross-session memory | Tasks, decisions, and files indexed per role across sessions |
| Role-based injection | Memory routed by agent role — each teammate gets relevant context |
| Auto-learn | /brain-learn extracts 15+ conventions from your git history |
| Solo mode | Full token savings and memory without Agent Teams |
| Fully local | SQLite + FTS5, no cloud, no telemetry, zero external Python dependencies |
| Cross-platform | macOS, Linux, WSL2, native Windows — all hooks run via Python |
Architecture
All data is local in ~/.claude-teams-brain/projects/<hash>/brain.db (SQLite + FTS5).
8 lifecycle hooks capture everything → role-based memory → ranked + deduplicated → injected within a 6000-char context budget.
For full technical details, MCP tool reference, and troubleshooting, see the full documentation.
License
MIT
