PlanDB
The issue tracker your AI agents are missing
Start new thread
Santosh Radha

PlanDB - The issue tracker your AI agents are missing

by
Agentfield
•
Claude Code has a to-do list. Codex has a to-do list. Gemini has a to-do list. None of them have a planner. PlanDB replaces flat checklists with a compound dependency graph. Your agents decompose tasks, parallelize independent work, adapt plans mid-flight, and persist knowledge across sessions. The result: 10x more complex projects completed autonomously. One agent built a 3,769-line GPT-2 in Rust - 20 tasks, 7 experiments, zero human intervention. Open source · Rust · SQLite · Zero config

Add a comment

Replies

Best
Santosh Radha
Maker
📌
Hey PH! I'm Santosh, building PlanDB at AgentField. I started where most of you probably are, using GitHub Issues to tell Claude Code what to do. Create issues, organize them, let the agent pick them up. It works... until you're running 4-5 agents on the same codebase. They duplicate work, start coding before dependencies exist, and lose all context between sessions. I was playing air traffic controller via markdown files and issue labels. That's when I realized: the problem isn't the tool, it's the paradigm. GitHub Issues, Linear, Jira - they were designed for humans planning at human speed. Agents are fundamentally different: - They decompose mid-flight (a 6-task plan becomes 20 as they learn) - They need atomic claiming (two agents can't grab the same work) - They pivot entire subtrees when an approach fails - They need knowledge to surface automatically, not via search - They operate at machine speed, seconds, not sprint cycles None of that maps to a kanban board. Agents need a compound dependency graph, not a task list. So I built PlanDB: a task graph database designed for how agents actually work. Single Rust binary, SQLite-backed, zero infrastructure. Three interfaces: CLI for shell agents, MCP server for Claude Code/Cursor, HTTP API for custom setups. Install in 10 seconds, agents start using it immediately. What surprised me: agents don't just get more organized with PlanDB, they get dramatically better at execution. When the graph tells them what's independent, they naturally parallelize. When context auto-surfaces, they stop rediscovering things. Everything is open source (Apache 2.0). Would love to hear: - Are you hitting similar coordination pain with your agents? - What does your current agent workflow look like? GitHub: github.com/Agent-Field/plandb