Your fund’s AI agents

Agent limits and controls

Each agent is its own identity in your organization, with a system prompt, model settings, knowledge, teammates and channel bindings.

Your system prompt outranks chat messages

Platform safeguards come first, then your system prompt, then an Automation’s trusted instructions. Chat messages, Slack messages, knowledge pages and webhook JSON are data to reason about.

Up to 10 text sources per agent

Each source holds up to 500 KiB of valid UTF-8 plain text, Markdown, JSON, YAML, XML or CSV. HTML, PDFs, images and binary files are rejected.

Fetched sources serve website chat only

In Slack, an agent works from the conversation. An Agent message run works from its trusted instructions and one JSON object.

Revoked keys stop working within 30 seconds

A missing, unknown or origin-mismatched embed key fails closed. A revoked key stops resolving within 30 seconds at most. Allowed origins restrict browser embedding and don’t replace visitor authentication.

Disable an agent or retire it

Organization admins with permission to manage agents look after each one. Disabling pauses an agent and keeps its setup. Retiring asks you to type its name, removes it from every room and team, and keeps its message history.

No automatic reposts to Slack

An Agent message is saved before its Slack post is claimed. After a claim, Infra One never posts again on its own. Run history asks you to check Slack first.