By montinou
Universal self-learning and agent coordination for Claude Code. Captures trajectories, learns patterns, shares knowledge across projects.
Beta-Bernoulli confidence tracking with empirical Bayes cold-start and exponential forgetting. Use when building pattern reliability scores from noisy reward signals, cold-starting new items with partial pooling, or implementing temporal decay in non-stationary bandit systems.
Hierarchical Thompson sampling with cluster-level posteriors + 10% exploration + SNIPS counterfactual updates. Use when building retrieval/recommendation systems with implicit feedback that need to balance exploitation with exploration at scale (10k+ items).
Anti-bloat curation for learned knowledge bases — quality gates at ingestion, cosine-based dedup with LLM verification, credible-interval retirement, temporal staleness. Use when building self-updating RAG/pattern stores that must stay small and high-signal.
Trigger immediate pattern consolidation from recent trajectories. Use when asked to run learning, consolidate patterns, or process recent agent logs.
Pairwise LLM-as-judge evaluation with position randomization, active learning, and cost control. Use when building offline ground-truth labels for retrieval/recommender systems where explicit user feedback is unavailable or too sparse.
Admin access level
Server config contains admin-level keywords
Executes bash commands
Hook triggers when Bash tool is used
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Modifies files
Hook triggers on file write and edit operations
Modifies files
Hook triggers on file write and edit operations
Persistent memory, autonomous learning, and agent coordination for AI-powered development
Quoth is a two-tier system:
Cloud Platform (src/) Plugin (quoth-plugin/)
├── Next.js 16 + React 19 ├── 22 MCP tools (stdio)
├── Neon PostgreSQL (Drizzle ORM) ├── SQLite + HNSW (local)
├── Clerk authentication ├── Background daemon
├── Vercel AI Gateway embeddings ├── Trajectory capture
│ (text-embedding-3-large, 2000d) ├── Pattern learning (Haiku)
├── Cohere reranking ├── Intelligence routing
├── Cloud MCP server (/api/mcp) └── Hook-based automation
└── Dashboard + API routes
Available via HTTP at https://quoth.triqual.dev/api/mcp:
| Tool | Description |
|---|---|
quoth_search_index | Semantic search with text-embedding-3-large (2000d) + Cohere reranking |
quoth_read_doc | Retrieve full document content by ID |
quoth_read_chunks | Retrieve document chunks for granular access |
quoth_memory_store | Store a memory entry for the current agent/session |
quoth_memory_search | Semantic search over stored memories |
quoth_memory_list | List stored memory entries |
quoth_memory_forget | Delete a memory entry |
quoth_agent_register | Register an agent in the project |
quoth_agent_list | List registered agents |
quoth_agent_assign | Assign an agent to a task |
quoth_agent_send_message | Send a message to another agent |
quoth_agent_inbox | Read an agent's inbox |
quoth_agent_tasks | List tasks assigned to an agent |
quoth_agent_task_reassign | Reassign a task to a different agent |
quoth_project_create | Create a new Quoth project |
quoth_project_invite | Invite a collaborator to a project |
quoth_token_generate | Generate an MCP access token |
quoth_genesis | Bootstrap project documentation (minimal/standard/comprehensive) |
Available locally via stdio (quoth-learning server):
Patterns (8): quoth_log_outcome, quoth_score_pattern, quoth_top_patterns, quoth_search_patterns, quoth_project_patterns, quoth_promote_global, quoth_seed_from_exolar, quoth_propose_update
Agents (6): quoth_daemon_status, quoth_ingest_trajectory, quoth_agent_register, quoth_agent_heartbeat, quoth_agent_list, quoth_assign_task
Intelligence (6): quoth_route_task, quoth_intelligence_init, quoth_intelligence_context, quoth_intelligence_consolidate, quoth_intelligence_stats, quoth_intelligence_feedback
Skills (2): quoth_extract_skill, quoth_list_skills
All hooks run through a unified dispatcher (hook-dispatch.js). Zero API calls in automatic hooks.
| Hook Event | What It Does |
|---|---|
UserPromptSubmit | Route task to optimal agent, show relevant patterns |
SessionStart | Init intelligence graph, inject top patterns (>= 0.6 confidence) |
SessionEnd | Consolidate intelligence graph, recompute PageRank |
PreCompact | Same as SessionEnd (pre-context-compression) |
PostToolUse (Write/Edit) | Record edit for intelligence |
PostToolUse (Bash/Write/Edit/Agent) | Capture tool calls to trajectory file |
PreToolUse (Bash) | Block dangerous commands |
SubagentStart | Inject domain-relevant patterns via additionalContext |
SubagentStop | Implicit positive feedback to intelligence |
claude mcp add --transport http quoth https://quoth.triqual.dev/api/mcp
cd quoth-plugin && bash scripts/setup.sh
Symlinks hooks to ~/.quoth/hooks/, injects hook declarations into ~/.claude/settings.json, and adds MCP server config. Idempotent.
| Variable | Description |
|---|---|
CLERK_SECRET_KEY | Clerk authentication secret key |
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY | Clerk publishable key (client-side) |
DATABASE_URL | Neon PostgreSQL connection string |
AI_GATEWAY_API_KEY | Vercel AI Gateway — text-embedding-3-large (2000d) |
JWT_SECRET | MCP token generation |
RESEND_API_KEY | Email delivery (optional) |
QUOTH_API_KEY | Plugin cloud sync (qth_* key) |
npm run build # Build Next.js app
npm test # Run Vitest tests
npm run lint # ESLint
Unified test automation plugin integrating Playwright execution, Quoth patterns, and Exolar analytics. Ad-hoc testing, ticket-to-test workflows, and self-healing tests.
npx claudepluginhub montinou/quothPersistent memory for AI coding agents -- captures tool usage, compresses via LLM, injects context into future sessions. 12 hooks, 41 MCP tools, 4 skills, real-time viewer.
memX: local-first semantic memory for coding agents. Native Claude Code lifecycle hooks.
Skill memory layer for Claude Code — auto-capture, learn, and reuse skills from Acontext
Local-first personal AI identity layer for MCP-compatible coding tools. Stores your lessons, decisions, playbooks, and project context as local JSON. AI proposes; high-risk items wait for your review, and everything stays visible and reversible. Local-first, no cloud, no account.
Stateful agent tools for Claude Code. Learns who you are, remembers what you're working on, schedules tasks, maintains itself.
Persistent memory for Claude Code — memories survive across sessions, projects, and machines