Suggests plan edits
When your numbers or the research change, it proposes the edit as an inline diff under the section. You accept or reject it, and version history can undo an accept.
The co-founder
It edits the plan, runs the board, and reads Stripe, GitHub, Sentry, PostHog and Slack. On Plus and Pro it can also work on your code in a private VM. It asks before anything leaves the app.
One agent
The same co-founder in the sidebar and in chat. It reads your plan, board and connected tools, and asks before anything leaves Founder2.
When your numbers or the research change, it proposes the edit as an inline diff under the section. You accept or reject it, and version history can undo an accept.
It adds tasks, projects and documents, and turns research findings into follow up tasks. You can move or delete any of it.
On Plus and Pro it pulls Stripe, GitHub, Sentry, PostHog and Slack into the plan as live numbers, and the Stats page keeps a daily history.
Jobs
You keep one conversation with the co-founder. Bigger work becomes a background job it starts for you.
Your message starts streaming at once. A plan edit arrives as a suggestion to accept or reject, and a board change happens in the thread.
It searches the web, reads the top pages, writes a research document and adds follow up tasks. You approve the job first, track it on the Co-founder page, and can cancel it while it runs.
Web pages, issue text and replies from your other agents are content to read, never instructions the co-founder follows.
Opening a GitHub issue, adding a batch of tasks or messaging another agent shows you the exact payload first.
A daily brief or a co-founder task runs at the hour and time zone you set, and the result is saved in the app.
Approvals
Before you approve, the card shows the exact payload: the issue body, the task list, the message. The button says what will happen.
Reading your tools and the web, and changing your own board. Plan edits are suggestions: nothing in the plan changes until you accept the diff.
Opening a GitHub issue, commenting on one, starting a research job, adding a batch of tasks, marking a task verified, and the first message to another agent of yours.
For the riskiest actions, such as sending email or merging code. Founder2 does not do either today, so there is nothing to confirm yet.
Your data
You paste restricted keys. Founder2 stores them encrypted, and only reads with them.
Stripe, Sentry, PostHog and Slack are only read. The one outside write is a GitHub issue or comment that you approve.
Integration keys are encrypted in the database and are not shown again after you save them.
Web pages, issue bodies and replies from other agents are content to analyse, never instructions the agent follows.
An approval runs once. A double click or a second tab cannot run it twice.
Agent VM
On Plus and Pro the co-founder can start a private Agent VM with your repo. A coding agent reads the code, edits files and runs your tests. It cannot push or open pull requests yet.
It stays asleep until a job needs it, and pauses when it is idle.
The coding agent cannot push, open a pull request or reach the network. Pull requests are coming soon, and each will wait for your approval.
VM time is not counted or capped per plan yet, so there is no hours allowance to quote.
Memory
A new message does not start from a blank prompt. The plan, the board and your memory notes are loaded into the turn.
Every turn receives an outline of the plan, the board and your notes. Edit the plan and the next turn sees it.
Ask it to remember a fact, a preference or a decision. Each note is listed on the Co-founder page under What I remember.
Open the workspace on another device and the conversation is there. Switching the model keeps it.
Nothing here fine tunes a model on your company. The record is the plan, the notes and the thread, and you can export all of it.
Reach
Other agents can work with the co-founder today. It runs on the open model you pick in Settings, served through OpenRouter: Kimi, DeepSeek, Qwen, GLM, MiniMax, Muse Spark Open and others.
Connect another agent over MCP or A2A, or Poke. The co-founder can message it once you approve the first message.
Founder2 runs an MCP server and an A2A endpoint, so another agent can read your plan or ask the co-founder a question, with a token you create.
Ask a question or approve a decision from a chat app.
Open Settings in your workspace, then MCP access. Create a labelled token for your agent and keep it private.
In your agent’s MCP settings, use your Founder2 origin followed by /api/mcp. Supply the token as an Authorization: Bearer header.
Ask your agent to read your plan or tasks. Revoke its token in Settings whenever you need to cut access.
Codex, OpenCode and Cursor connect through their MCP configuration. Your Founder2 token controls access to your workspace; it is not a model API key.
Open your workspaceOn the way
None of these exist in the product today.
Flags contradictions, claims with no source and lines that no longer match live data.
Joins your tools into one chain that ends at a cause, such as a cancellation spike and the error behind it.
A practice sales call with a buyer built from your own research, labelled as a simulation.
The coding agent opens a pull request from the Agent VM. You merge it.
Plus, at $29 a month, adds live data and the Agent VM. The first 14 days are free.
Start your 14 day trial