Workflow state for AI coding agents

Token-saving durable context for AI coding agents

djobs gives Codex, Claude, Gemini, Copilot, Cursor, Cline, and any MCP-compatible coding agent durable, resumable task memory. It saves every step of long-running work as workflow state, so your agent resumes cleanly after crashes or context loss - nothing lost, nothing redone, fewer tokens spent replaying finished work.

Install the VS Code extension $ djobs init
✓ MIT licensed ✓ Any MCP agent ✓ SQLite — zero infra ✓ Token savings estimate
Codex Claude Code Gemini Cursor Cline Copilot MCP server AI coding agents resumable workflows workflow state
Animated terminal demo: djobs resumes the unfinished tasks after an IDE crash, with no files lost and none re-done.
Built on MCP
  • Tested: Copilot in VS Code
  • Claude Code
  • Codex
  • Gemini
  • Cursor
  • Cline
  • GitHub Copilot
  • Any MCP client

Why djobs

Long agent tasks shouldn't vanish when the IDE does.

When a multi-file refactor or migration is interrupted, the agent's progress normally lives only in chat history — and disappears with it. djobs makes that progress durable, inspectable, and resumable.

Local-first memory

Tasks, status, and a full audit trail live in SQLite by default. No broker, no daemon, no cloud lock-in.

Agent-native MCP

Durable enqueue, complete, inspect, and resume tools are exposed directly to your agent over the Model Context Protocol.

Human-inspectable

See exactly what remains, what succeeded, what changed, and estimate how many replay tokens durable evidence saved.

How it works

Enqueue, crash, resume — nothing lost, nothing redone.

djobs also installs agent instructions, so coding agents proactively call resume_session, enqueue_task, complete_task, and fail_task during long or risky work — you don't have to remember to ask.

Agent calls enqueue_task

As it plans a long job, the agent saves each unit of work as a durable task with a clear summary — a checkpoint that survives any crash.

The IDE crashes

Context resets, the chat disconnects, the window closes. The in-flight plan would normally be gone — but every task is already on disk.

Agent calls resume_session

On the next session, the agent instantly recovers the unfinished tasks and continues — completing what's left without redoing finished work.

Get started

Two ways in. Both take a minute.

djobs is workflow state for your agent, not a dependency of your app - so it works in any repo. In VS Code, the extension installs and manages everything for you. For other MCP agents, one command wires the project. djobs installs its own runtime; you don't manage Python or config by hand.

  • installs runtime
  • wires MCP server
  • installs agent instructions
  • adds VS Code sidebar
# VS Code / Copilot:
# install extension -> run "Set up djobs"

# Codex / Claude / Gemini / Cursor / Cline:
djobs init
✓ runtime installed · MCP wired · instructions added · doctor OK

Give your agent a memory that survives the crash.

Open source, MIT licensed, and built for calm multi-agent work. One install gives every MCP-aware coding agent the same durable workflow state.

$ djobs init See setup