Commt: Private AI agents, scoped to what you allow

Hey Product Hunt,

I'm Mesut, one of the makers of Commt, a private AI agent platform for businesses. You build and configure AI agents around your own knowledge, tools and workflows, and each one only reaches what you've explicitly allowed.

An agent that reads your documents or touches internal systems needs someone deciding what it can reach, not guessing after the fact.

Every agent runs inside four boundaries:

- Knowledge: only the documents and knowledge bases you attach, retrieval-augmented generation is optional rather than assumed
- Tools: only the connectors and MCP servers you scope to that agent, none by default
- Network: web search and page browsing are on by default, and you switch them off per agent
- Retention: choose how long conversation history is kept, from nothing up to a year

Twenty tool connectors are available at launch, from Slack and Microsoft Teams to Jira, Notion and Salesforce. Custom MCP servers run in an isolated sandbox, where they are checked for security issues before they can be used.

Each plan carries a fixed monthly request allowance instead of a metered bill, so usage stays predictable rather than turning into a surprise invoice.

Early access pricing: Entry and Pro are 30% off, locked for 12 months, until 29 September, 07:00 UTC.

A couple of differences worth naming against tools you might already use:

- Compared to Dust: Commt checks custom MCP servers for security issues in an isolated sandbox before use, with a fixed monthly request allowance instead of a per-seat plan and credit pool that varies by model and task. Dust also scopes tools per agent, through what it calls Spaces.
- Compared to Botpress: retention can be set to nothing at all, where Botpress's documented default is 90 days for conversations and messages. Botpress also offers per-bot knowledge bases.

I'd love to know: what's the one system or document set you'd want an agent to reach first?

Thanks for checking it out.