We posted Tako AI here last week (open-source AI agent for Okta). We just shipped the Slack integration.
GitHub: https://github.com/fctr-id/okta-ai-agent
The enthusiasm for AI agents living in chat apps is massive right now. The workflow benefit is obvious: you don't leave the context where you're already working.
But for many of these tools, security feels like an afterthought — exposed local ports, permissive defaults, and unchecked execution. When the data involved is Okta user records, admin group memberships, and app entitlements, that model doesn't work. Security can't be a second-class citizen.
So Tako's Slack bot ships locked down by default. Nobody gets access until you explicitly add their Slack ID to the .env whitelist. And your Okta data never touches Slack's servers — the bot runs locally, queries your own infrastructure, and posts results back to the thread.
All commands:
/tako [question] → ask anything about your Okta tenant in plain English
/tako history → your last 5 queries with ▶ Run and Star buttons
/tako favorites → your starred queries, always one click away
/tako help → full command reference
Two implementation details worth sharing:Socket Mode by default. Tako opens an outbound WebSocket to Slack — no port forwarding, no reverse proxy, no public URL. Most identity tooling runs on internal networks that can't expose HTTP endpoints.
Re-auth on every action. We re-check authorization on every interactive button click, not just the initial slash command — because Slack action payloads can be replayed by anyone who intercepts them.
Happy to answer questions on the implementation.
—Dan