What Celemine for Slack does
Celemine is a governance layer between your own AI agents and the systems they work in. It acts as a circuit breaker for agents that use tools such as Slack, GitHub, Google Workspace and HubSpot. Anomalous activity is stopped before accidents happen.
Connecting Slack lets the agents you authorize work in Slack on your behalf. They can search messages, read channels and conversations you have access to, and post as you. They get no access beyond what you already have in Slack.
How it works
- Add Slack to Celemine. Authorize Celemine in Slack and approve the permissions below. You choose whether agents can reach private channels and direct messages.
- Set your boundaries once. Write plain-language rules for what agents may search, read and post, or link them to your company policies.
- Connect your agents. Point Claude, OpenAI, Gemini, Mistral or another agent harness at Celemine. Celemine is the MCP gateway, so Slack is one of the apps your agents reach through it.
- Every call is checked. Agents declare their intent in a Session. Each Slack call must pass policy, your boundaries and a check against the Session's purpose before it runs.
If a reviewer confirms a boundary breach, the Session closes and the agent is suspended until a person releases it from the Celemine Inbox.
Get started
Add Celemine to Slack, then follow the steps in Celemine to connect your first agent. After you approve the permissions, you will land on a page that confirms Slack is connected and shows your next steps.
You do not need to map accounts by hand. You authorize Celemine in Slack, and Celemine links your Slack user to your account.
Permissions we request and why
Celemine asks for the smallest set of permissions that lets agents search, read and post. Everything below is a Slack user permission, which means Celemine acts as you, for agents you have connected, and only after its checks approve the call.
Always requested
| Permission | How Celemine uses it |
|---|
search:read.public | Lets an agent you have connected search messages in public channels you can access, after Celemine checks each search against your policy and boundaries. |
|---|
search:read.users | Lets an agent find workspace members by name, so it can address the right person in a message you have allowed it to write. |
|---|
channels:history | Lets an agent read message history in public channels you can access, only within an approved Session. |
|---|
channels:read | Lets an agent list public channels and their details, so it can find the right channel before Celemine approves a read or a post. |
|---|
chat:write | Lets an agent post as you, only when Celemine's policy, boundary and interpretation checks approve the call for that Session. |
|---|
users:read | Lets Celemine and the agents you connect look up workspace members' names and profiles. Celemine uses it to attribute each action to you, and agents use it to address people correctly. |
|---|
Optional: you choose at install
Celemine works with the permissions above alone. These optional permissions extend agents to private conversations. You can decline them and add them later by reconnecting Slack from Celemine.
| Permissions | What you unlock |
|---|
search:read.private, groups:history, groups:read | Search, list and read the private channels you belong to. Without them, agents can only reach public channels. |
|---|
search:read.mpim, mpim:history, mpim:read | Search, list and read your group direct messages. Without them, agents cannot see group DMs. |
|---|
search:read.im, im:history, im:read | Search, list and read your direct messages. Without them, agents cannot see your DMs. |
|---|
Respecting Slack's privacy model
- Agents only reach content that you can already access in Slack. Private channels and direct messages are included only if you opt in.
- Messages posted by an agent are posted as you, so you stay accountable for them. Boundaries and approvals let you control what agents may post.
- Celemine does not use Slack data to train AI models.
- Credentials in agent plans are refused, and values that only need matching are hashed.
AI components and data handling
Celemine uses AI reviewers to judge whether an agent's call fits its declared purpose. What a reviewer sees of a call is sanitized by default.
- AI reviews run with European providers under zero data retention, so the sanitized content of agent requests is not retained by the model provider.
- All customer data is stored in EU data centers.
- Session plans and mandates are deleted after 15 minutes.
- Celemine does not make consequential decisions about people. A person releases a suspended agent from the Inbox.
AI outputs can be inaccurate. Agents you connect, and the summaries or messages they produce, may contain mistakes, so review anything important before relying on it. Read the full list of subprocessors in our privacy policy.
Security
Customer data is encrypted in transit with TLS 1.2+ and at rest with AES-256. Each organization and each agent runs in a segmented environment with no access to other agents' secrets. Learn more on our security page.
To report a security issue, email hello@celemine.com.
Privacy
Read our privacy policy for what we collect, how we use it, how long we keep it, and how to request access, transfer or deletion of your data. You can also email hello@celemine.com to make a request.
Support
Email hello@celemine.com for help. You do not need an account to get support, and we aim to reply within two business days.
Disconnecting Slack
You can disconnect Slack from Celemine at any time. You can also remove Celemine under Apps in your Slack workspace settings. Either one revokes the access tokens Celemine holds for you.