Distributed, git-backed graph issue tracker for AI agents. “Wake up, ask what’s next, the agent knows”
-
https://steve-yegge.medium.com/introducing-beads-a-coding-agent-memory-system-
-
provides a persistent, structured memory for coding agents. It replaces messy markdown plans with a dependency-aware graph
-
version-controlled SQL database
-
JSON output, dependency tracking
-
hash-based IDs to prevent merge collisions in multi-branch workflows
-
memory decay compacts/summarizes old tasks to save context
-
graph links for knowledge graphs -
relates_to,duplicates,supersedes
problem statement
- vibe coding currently requires constant mundane monitoring.
- Dementia problem: Coding agents have no memory between sessions
solution
- “Trapping the Devil”: Centralize a plan store
- attempted using Markdown files that referenced each other but got very messy
- pivot to issue tracker
- built a SQL-based issue tracker with a CLI for creating and updating issues
- helps agents stay on track with long plans and shifting priorities
- can kill agents after they complete each issue - Beads helps agents find where to pick up again
- acts like a managed database but writes issues into Git as JSON lines, allowing version control - AI does intelligent merging when conflicts are encountered
- naturally distributed- worker agents can work on the same project in parallel, sharing the same database backed by Git in the same project as your code
- instead of scanning a Markdown to-do list, can just run CLI command and see definitive list of unblocked work
- solves multi-agent coordination - two agents can query the same database and see what’s claimed or ready