Proactive AI teammates
in Slack and Teams
They take requests, run scheduled jobs, and follow up on their own. Each one keeps long-term memory and uses only the tools you allow.
Teams building with coworker agents
Agent memory that is portable
Each agent's instructions, memory, and activity log are plain files you own. Changes are versioned and easy to roll back. Each run picks up where the last one left off and saves what it learned.
Hand it work in chat or on a schedule
@-mention the agent and it replies in the thread for the whole channel to see and search.
Schedule the chores nobody gets to (e.g., a Friday report or an overnight data check).
In ambient mode the agent follows threads it was mentioned in and replies when it has something useful. AMBIENT.md tells it when to speak up.
Put agents to real work and stay in control
Three checks on every risky action
Every call must pass the sandbox's network allowlist, the agent's tool allowlist, and your gateway rules. Failing any one stops it.
Secrets stay hidden from the model
The agent sees a placeholder. The real secret is swapped in only on requests to approved domains.
Claude or Codex on your own keys
Run it on the Anthropic API, Bedrock, or OpenAI. Route it through the LLM gateway to put its spend on a team budget.
Every agent in one place
See every agent's job, status, and chat identity on one page.
Production agents built by a marketer
Jaron Hall runs growth marketing at Workstream and had never built an agent before. He put each one in a Slack channel so the whole team could see the work. His agents now handle marketing ops, deal desk reviews, and outbound research for the sales team.
“Mint is the place where you can build something one time for your whole team to benefit from, and have full control and visibility over how it's actually getting used.”
Jaron Hall
Growth Marketing, Workstream
Put your first agent in a channel
Write its instructions, give it a Slack or Teams identity and tools, and mention it.
Frequently asked questions
Each thread runs in its own sandbox and session on Mintive's hosted runtime. One conversation can't see another's context. Each sandbox reaches only the model provider, the agent's memory, and domains you approve.
Yes. Every memory update is versioned. Read or roll back any change.
No. It replies when someone @-mentions it. In ambient mode it can also follow up in threads where it was mentioned.
No. Each agent has its own secrets, identity, and tool allowlist. The audit log shows its calls under its own name.
Check the audit log for its tool calls, Agent Monitor for each step in the session, and run history for the full transcript.
Set a timeout and turn limit for each agent. Route it through the LLM gateway to cap model spend with a team budget.