Shared project memory for the agents on your team.
Memtack is where a software team keeps project context, decisions, notes, diagrams, and data models. Coding agents read that record over MCP, so a new session does not start from an empty chat. A person reviews tasks, and the work ties to a GitHub pull request.
Memtack is where a software team keeps project context, decisions, notes, diagrams, and data models. Coding agents read that record over MCP, so a new session does not start from an empty chat. A person reviews tasks, and the work ties to a GitHub pull request.
What gets lost between sessions
- Agents forget the project's context between sessions.
- Decisions end up scattered across chats and separate docs.
- Bringing a new person, or a new agent, onto the project is slow.
How a session starts
Three steps. The order matters.
1
Keep the record
Future screenshot: context, decisions, notes, diagrams, and data models kept with the project. The team writes context, decisions, notes, diagrams, and data models in Memtack.
2
The agent reads it
Future screenshot: the agent retrieving the shared record over MCP at the start of a session. At the start of a session the agent reads that record over MCP, instead of relearning the repository from scratch.
3
A person reviews the task
Future screenshot: a person reviewing the task. The work is linked to a GitHub pull request. Memtack does not open or review that request by itself. Someone on the team reviews the task. The work ties to a GitHub pull request.
What the record holds
Only the things a team already has to keep somewhere. Nothing else is claimed here.
Memtack is read over MCP. Cursor, Claude Code, and Codex are the clients named for the pilot. Other coding agents can connect the same way, through MCP.
Context and decisions sit in separate chats and docs. A new session starts without that record.
Future screenshot: context and decisions spread across chats, before they live in one record.
The same facts stay with the project: context, decisions, notes, diagrams, and data models.
Future screenshot: the record holding context, decisions, notes, diagrams, and data models together.
Cursor, Claude Code, and Codex are the clients named for the pilot. Other coding agents can read the same record over MCP.
Future screenshot: an agent reading the shared record over MCP. The agent does not open or review pull requests by itself.
A person reviews the task. The work ties to a GitHub pull request. That review stays with a person.
Future screenshot: a person reviewing the task, with the work linked to a GitHub pull request.
- Context
- The working facts of the project, kept with the project.
- Decisions
- What the team decided, in the record rather than in a chat.
- Notes
- Notes written once, where the next person and the next agent can read them.
- Diagrams
- Drawings of the system, stored with the rest of the record.
- Data models
- The project's data models, described in the same place.
- Tasks
- A person reviews the task. The work ties to a GitHub pull request.
What this site stores
The product is not open for accounts. This marketing site stores the email and the language of the page. The form also asks for team size and which agents you use. Those answers are checked and not stored. An IP address is held in memory for a short time to limit repeated submissions.
That list is for contacting teams about the pilot. It is not sold. A full privacy policy will be published before any pilot starts.
Team pilot
$49 a month
For a software team of about 3 to 15 people during the pilot.
Payments are not connected. Polar, the payment provider planned for the pilot, is not set up.
Every link and button on this page joins the waitlist. You will not be charged.
Questions
Join the waitlist
The team pilot is $49 a month. Payments are not connected. This form does not charge you.