Notion
Project intent, work specifications, decisions, run history, and durable checkpoints.
SOURCE OF TRUTH FOR WHY + WHATSystem 001 · launch edition
Agent Project OS
Persistent project memory, an agent-ready work queue, and a complete activity ledger for developers working with AI coding agents.
$49One-time purchase · launch edition updates included
Built for Notion users working with MCP-connected agents. No proprietary app or subscription required. The sandbox checkout grants immediate access to the public Notion template for end-to-end testing.
PROJECTAgent Project OS
Context healthyREADY04
RUNNING02
REVIEW03
DECISIONS17
WORK ITEMAGENTSTATUS
Add project checkpoint protocolCODEXRUNNING
Test MCP handoff workflowCLAUDEREVIEW
Define queue claim semanticsHUMANREADY
MISSION CONTROLProjects · work items · runs · decisions · checkpoints
The problem
AI sessions are disposable. Your project memory should not be.
Every new agent session starts by reconstructing the same context: what the project is, why earlier decisions were made, what changed, what failed, and what should happen next.
A clear division of responsibility
Notion stores the intent. Git stores the code. Agents do the work.
Agent Project OS connects the three without pretending Notion should replace your repository or your coding tools.
Agents
Load context, claim ready work, execute within boundaries, verify results, and report back.
DISPOSABLE EXECUTION SESSIONSGit
Source code, tests, commits, pull requests, releases, and the technical record of change.
SOURCE OF TRUTH FOR CODEInside the system
Seven databases. One durable project brain.
The initial architecture is deliberately small enough to use and structured enough to support real agent workflows.
Projects
Canonical objective, repository, working rules, architecture, current milestone, and latest checkpoint.
Work Items
Agent-ready features, bugs, investigations, reviews, and maintenance tasks with precise completion criteria.
Agent Runs
One record per attempt: agent, plan, files, commands, verification, result, blocker, and handoff.
Activity Events
An append-only chronological history of meaningful project events without low-level tool noise.
Decisions
Context, alternatives, rationale, consequences, authority, and superseded decisions.
Checkpoints
Compressed snapshots that let a new session understand current project state without reading everything.
Agents
Capabilities, instructions, permissions, suitable work, prohibited actions, and review requirements.
Interactive instrument
Would an agent know what “done” means?
Agent Project OS turns vague tasks into executable work. Toggle the fields to see how the readiness score changes.
Task readiness50%
Needs more context
Interactive lab
Four tools for safer agent work.
Each instrument is a self-contained HTML file. It runs locally in the browser, sends no data anywhere, and keeps working without a model call or external JavaScript dependency.
01↗02↗03↗04↗
Task Readiness Checker
Find missing goals, boundaries, criteria, verification, and authority before an agent starts.
Autonomy Risk Calculator
Match an agent’s autonomy to the production, data, security, financial, and rollback exposure.
Context Health Score
Audit whether a fresh agent can understand the project without reconstructing weeks of history.
Agent Handoff Builder
Create a portable Markdown packet containing state, evidence, open problems, decisions, and next action.
The execution protocol
A repeatable handoff from human intent to verified work.
The included protocol tells agents how to load context, validate tasks, preserve decisions, and stop when authority is missing.
- 01Load project state
Read the project brief, latest checkpoint, selected work item, and linked decisions.
- 02Validate and claim
Confirm the task is executable, dependencies are complete, and the work fits the agent’s authority.
- 03Create a run
Record the plan, expected changes, risks, and verification strategy before changing code.
- 04Execute and verify
Perform the work inside the documented scope and run the specified checks.
- 05Report and preserve
Update the result, log meaningful events, propose decisions, and leave the next agent better informed.
What you receive
More than a duplicated workspace.
Complete operating system
Seven connected databases, filtered dashboards, templates, realistic example data, and a guided Start Here page.
Agent instruction pack
A canonical execution protocol plus ready-to-adapt instructions for planning, implementation, review, and testing agents.
Practical examples
Feature implementation, bug investigation, failed-run recovery, project resumption, and cross-agent handoff.
Interactive tools
Task specification checker, autonomy risk calculator, context health score, and handoff builder as they are released.
Setup and operating guide
MCP connection, minimum permissions, first project setup, first agent run, review, checkpoints, and recovery.
Launch edition updates
Improvements throughout the launch edition as the system is dogfooded and refined with early customers.
Honest boundaries
Built for controlled collaboration, not autonomous chaos.
Version one is for
- One person and one or several projects
- Human-directed or agent-pull workflows
- One active implementation agent per project
- Structured handoffs and human review
- Persistent decisions and checkpoints
Version one is not
- A replacement for GitHub, Linear, or Jira
- A distributed multi-agent job queue
- An automatic deployment system
- A place for passwords or production secrets
- A promise of fully autonomous development
Questions
Before you buy.
Which Notion plan do I need?
The core system uses standard pages and databases. Optional native agent or advanced automation features may require a higher Notion plan; those are not required for the base workflow.
Which AI tools does it support?
The architecture is tool-independent. It is designed for AI tools that can read and update Notion through MCP or an equivalent integration. Tool-specific instructions will identify what has actually been tested.
Does this run agents automatically?
No. The launch edition gives agents structured context and a clear operating protocol. You remain in control of when agents run and which work they may perform.
How will I receive it?
Polar handles checkout and access. After purchase, your customer portal reveals the Agent Project OS Notion link. Open it and click Duplicate to copy the complete workspace into your own account.
Is this an official Notion product?
No. Agent Project OS is an independent product created by Flavio Copes and is not affiliated with or maintained by Notion Labs, Inc.
Launch edition · $49
Give the next agent a project that remembers.
One system for persistent context, executable tasks, agent runs, decisions, and handoffs.
Open sandbox checkoutPolar sandbox checkout for end-to-end testing. No real payment is processed.