Brigade

Brigade is a local control plane designed for managing multi-agent coding workflows, particularly where multiple AI coding agents operate together and require shared memory, tools, and configuration. It addresses coordination challenges that arise when agents each maintain separate configurations, skills, and session memories, often resulting in drift, unread notes, and silent failures. Brigade consolidates these elements into a single, reviewed source of truth within a repository, ensuring that all agents share a unified catalog of MCP servers, tools, and skills, and that memory handoffs are durable and accessible across sessions and agents. The tool operates without a background daemon or lock-in, and is installed as a Python CLI via pipx. Brigade enables agents to record file receipts for every run, capturing commands, real exit codes, and changes, which can be used to verify actions and promote skills only when validated by actual exit codes. It supports dry-run diffs before applying changes, and merges configurations by server key without overwriting manually added servers. Secrets are managed as variable references and are not inlined into configuration files. Brigade’s memory engine stores agent notes as knowledge cards in markdown files, with a slim index to keep memory manageable and within context window budgets. Notes are linted, classified, and routed for durability, with ambiguous entries awaiting human review. Memory can be served read-only over MCP, allowing agents to recall information mid-task. The platform also supports optional plugin stations such as GraphTrail for local code graphs, MiseLedger for evidence ledgers, Agent Pantry for encrypted session and secret sync, Content Guard for secret management, and Token Glace for compacting tool output. Built-in portable skills are managed and reconciled based on real workflow outcomes, not self-reported results. Brigade is aimed at engineers and developers orchestrating autonomous or multi-agent coding environments. It is distributed under the MIT license, has zero runtime dependencies, and is developed in Python. The tool’s modular design allows it to function independently or with additional stations as needed.