TaskSeam

Switch AI. Keep working.

TaskSeam keeps the accepted state of a project available when work moves between Codex, Claude Code, VS Code agents, ChatGPT, and Claude.

Website · Editor setup · Quick start · Demo · Privacy

Public alpha: The local CLI, MCP server, agent plugin, project state, provenance, and per-agent continuation are working. The browser extension is a developer preview.

Quick start

Install TaskSeam:

pipx install taskseam

Connect the supported agents installed on your computer:

taskseam setup

Enable TaskSeam in a project:

cd your-project
taskseam init

Now open the project in Codex for VS Code/Cursor, Claude Code for VS Code, or their terminal interface and start a new chat. TaskSeam supplies the project's accepted decisions, active constraints, unresolved questions, and changes the agent has not seen. Durable changes use the agent's normal tool-approval screen.

Start with this prompt:

Continue this project. What accepted decisions, constraints, and open questions should I know?

There are no routine TaskSeam commands after initialization.

What changes for the user?

Without TaskSeam:

1. Explain the project again.
2. Paste an earlier conversation.
3. Correct an outdated summary.
4. Repeat when switching tools.

With TaskSeam:

1. Start an agent in the project.
2. Continue the work.
3. Approve durable project changes when prompted.

TaskSeam stores project state locally under .taskseam/, which it adds to .gitignore.

Supported surfaces

See the editor setup guide for the exact UI steps and the integration guide for technical details.

Why this is different from shared memory

Shared memory retrieves potentially relevant facts. Session summaries describe what happened in one conversation. TaskSeam maintains the current state of a project, preserves what changed, and tracks what each agent has already received.

Claude discussion ─┐
ChatGPT decision ──┼─> current project state ─> unseen changes ─> Codex
Code and Git ──────┘            │                       │
                            provenance              per agent

The goal is a small, explainable continuation rather than a transcript dump.

Privacy model

• Project state stays on the user's computer by default.
• Agent writes use the host's approval flow.
• Browser capture starts only after an explicit user action.
• Browser results remain editable before saving.
• TaskSeam does not operate an analytics, advertising, or hosted synchronization service.

Read the privacy policy and security policy for the current alpha boundaries.

Try the complete flow

The five-minute demo starts with a clean directory, accepts a project decision in one agent, continues in another, and verifies that the first agent later receives only what changed.

Project status

The public roadmap tracks the current priorities:

1. Validate clean installation across supported operating systems.
2. Publish the Python package to PyPI.
3. Package and submit the browser extension.
4. Improve automatic task matching and correction.
5. Publish repeatable continuity evaluations.

Development

git clone https://github.com/TaskSeam/taskseam.git
cd taskseam
python3 -m venv .venv
.venv/bin/python -m pip install -e .
.venv/bin/python -m unittest discover -s tests -v

TaskSeam is licensed under Apache-2.0. Contributions are welcome through focused issues and pull requests. Read the contribution guide before making broad architecture changes.

Created and maintained by Sai Sandeep Kantareddy.