Convex-Backed & MCP-Exposed

The shared memory layer for AI agents.

Stop losing context between development sessions. Cerebro is a secure, developer-owned project wiki that lets different autonomous agents query, update, and transfer task knowledge instantly.

Workspace: cerebro-core-refactor
MCP: Online
Projects
cerebro
anvil-dashboard
Memory Modules
Wiki Index12 pages
Handoff Logs48 saved

Session Handoff: ClaudeCodex

ID: session_9f83a21 • Saved 5 minutes ago

ACTIVE_STAGE
Task Summary
API token flow integration completed

Exposed /api/mcp endpoint with bearer authentication. Next agent needs to write unit tests for the token verification middleware.

State Attributes
Affected Files:lib/auth.ts, api/mcp/route.ts
Schema Version:2.1.0
Token Permissions:read_write
The Memory Wiki System

Engineered for LLM execution flow.

Memory Database Ledgertable: memory_items
refactor-auth-mcpMODIFIED
initialize-docker-stackMODIFIED
add-project-spec-rulesCREATED
01 / Workspace Ledger

Project Memory Ledger

Consolidate your task updates, API design agreements, and architecture notes in a centralized database schema. Incoming agents read this ledger at the start of their run, eliminating cold-start information gaps.

02 / Authentication

Cryptographic MCP Security

Protect your workspace context. Cerebro issues granular, scoped bearer tokens for each agent instance. Define precise read-only or read-write permissions, keeping your project data isolated.

Access Tokensscope_config
Token Key
crb_live_••••••3a19ACTIVE
read_memory
write_memory
admin_dashboard
Transition Cycle

The context lifecycle.

Cerebro tracks task execution in four distinct steps, creating a continuous feedback loop between AI agent sessions.

Phase 01

Capture

Active agents summarize their finished tasks and next steps automatically to the ledger upon exiting their session.

Phase 02

Scope

Task summaries are verified against active project scopes and securely stored using Convex data schemas.

Phase 03

Retrieve

Incoming agents request project context via the local MCP server, loading recent changes into their context.

Phase 04

Continue

Agents start execution immediately with context already established, eliminating repetitive instruction loops.

Infrastructure

Self-Hosted Docker Deployment

Own your data. Run the Cerebro stack securely on your private VPS or local developer workstation.

01 / Reverse Proxy

Caddy Router

Routes incoming public traffic securely over SSL, proxying dashboard and API requests to localhost-bound dockerized containers.

02 / Backend Services

Next.js & Convex

Runs the private dashboard, token generation interface, and database mutations within isolated Docker Compose runtimes.

03 / Database Layer

SQLite Volume

Saves persistent state in SQLite via a simple host volume mapping, making backup, export, and migration straightforward.