Beads – A memory upgrade for your coding agent (github.com)

🤖 AI Summary
Beads (bd) is a lightweight, alpha-stage memory system for coding agents that replaces ad-hoc markdown plans with a graph-based issue tracker designed for long-horizon, agentic workflows. Drop bd into a repo (bd init or curl installer) and agents will create, link, and manage issues on your behalf using four dependency types—blocks, related, parent-child, and discovered-from—so agents can automatically record newly discovered work, compute "ready" tasks (no open blockers), and reliably execute multi-step plans across sessions. It’s explicitly agent-first: every command has a --json mode for programmatic use, and common workflows (bd ready, bd create, bd dep add, bd close) map neatly into agent loops. Technically, Beads achieves a centralized-database feel without a server by storing committed JSONL (.beads/issues.jsonl) in git as the source of truth, with fast local SQLite caches (.beads/*.db) per machine. bd export/import syncs between JSONL and SQLite; git handles distribution and an AI-assisted merge process resolves conflicts. This gives sub-100ms local queries, offline operation, multi-machine consistency, a full audit trail, and easy extensibility (add custom tables to the SQLite DB). For AI/ML practitioners building coding agents, Beads addresses agent "amnesia," improves reproducibility and auditability of agent actions, and simplifies distributed coordination—making agentic development more reliable without running a server. Note: alpha — expect API changes; use for internal/dev projects first.
Loading comments...
loading comments...