Your agent asks a human
only when stuck.
Your agent searches what your team already knows, then Orako asks the right domain owners only when needed. Replies from Slack, Discord or the dashboard land back in the agent, and resolved conversations become searchable team history.
AI didn’t kill communication.
It hid it.
Agents move fast, but the moment a real decision is needed, the humans who know are out of the loop. The cost shows up quietly.
Agents guess
Missing context, they assume and ship the wrong thing, confidently. You find out in review, or in prod.
Knowledge is trapped
The answer lives in one teammate's head, or 300 messages deep in a thread. Nobody finds it twice.
The meeting tax
So you book a call to unblock a bot. Multiply by every decision, every day, every project.
From a blocked agent
to a resolved answer
Six moves keep agents moving: search what your team already knows, ask the right people only when needed, then preserve the full answer for next time.
Search what your team already answered
Before interrupting anyone, the agent searches resolved conversations and curated entries in Orako. If a useful answer already exists, it keeps working immediately. Humans are only asked when history falls short.
How is refresh-token rotation handled?
It asks everyone who owns the domain
The agent asks about a domain, not a person. Orako sends the question to the teammates who own it: a card in their dashboard inbox and a DM in Slack or Discord. No @channel spray; only the people who can actually answer.
refresh token rotation?
→ domain · authA conversation opens where they already are
Orako keeps everyone on one durable conversation and mirrors it between the dashboard and their provider of choice. Answer in Slack or Discord, it shows in the dashboard. The first useful reply unblocks the agent; later replies stay attributed in the same thread.
Dana replied — rotate refresh tokens on re-auth.
How should refresh tokens rotate on re-auth?
Should the refresh token rotate on every use, or only on re-auth?
On re-auth only. Rotate-on-use breaks parallel requests.
And the grace window for the old token? How long?
Last one: do we revoke the whole family on reuse?
One thread, as many questions as it takes
Real decisions have follow-ups. The agent can push several questions into the same conversation, so context builds up in one place instead of scattering across new threads and pings.
The answer flows straight back to the agent
The moment Dana replies, Orako pushes it back into the agent’s flow, right where it stopped. No copy-paste, no re-prompting. It reads the answer and keeps building.
Rotate on re-auth only. 30s grace, revoke family on reuse.
How should refresh tokens rotate on re-auth?
Rotate on re-auth only. 30s grace, revoke family on reuse.
Consider 60s grace for slow mobile clients.
Resolved once, reusable next time
The conversation flips to Resolved and becomes searchable team history, with who said what and when. A later reply isn’t thrown away, it stays as an attributed follow-up, searchable next to the answer.
Everything to keep humans in the loop
Ask by domain, not by name
Tag people to domains (auth, payments, infra, UX) and every question reaches the right owners. Every reply stays attributed.
Answer anywhere
Slack, Discord, or the dashboard inbox. People reply where they already are. No new tool to learn.
Knowledge that compounds
Every answer is saved, searchable, and attributed, later replies are kept as follow-ups next to it.
Built for coding agents
Claude Code, Codex, and any MCP client connect in a minute: register the remote server, authorize in the agent, no token.
Escalation that never drops
Silence triggers a nudge, then an alert in your team channel. Questions never fall through.
Open source, self-hostable
The core is open source under AGPL, read it, run it on your own infra. Or let us host it for you.
Set up once, in the time it takes to read this
No migration, no new tool for your team. An admin invites everyone; each person registers the remote server and authorizes their own agent in a minute, no token.
Create your org
An admin signs up and adds teammates with the domains they own — backend, payments, UX, product. That's the map Orako routes against.
Everyone plugs the MCP
One command in Claude Code, Cursor or Codex, then authorize inside the agent — no token to paste. Non-devs who hold answers can reply from Claude Desktop or ChatGPT.
You're done
The next time an agent is blocked, it already knows who to ask. No workflow to configure, no new habit to learn.
Simple, per-seat pricing
Choose the hosted service or run the open-source Core on your own infrastructure.
Self-hosted
The whole core, AGPL-licensed. Clone it, run it, keep every conversation on your own infra.
Team
Pay only for the people who answer. 14-day free trial, no card.
Founder Lifetime
A perpetual offline license for one production instance. No subscription and no renewal.
Questions about how it routes
Why not just put this in our docs, git or tickets?
Your agent already reads those, it only pings a human when the answer isn't there. Git shows what changed, not why; tickets show scope, not the reasoning that got dropped in a meeting. Orako captures that answer the moment it's given and makes it searchable, so the next agent, and the next dev, never asks it twice. It doesn't replace your docs; it captures what docs never hold.
How does Orako pick who to ask?
By domain ownership, not by name. Everyone who owns the domain gets the question, and every reply stays attributed in one durable conversation.
What if two people answer?
The first useful reply can unblock the agent immediately. Later replies are not discarded; they remain attributed in the same conversation.
What happens if nobody responds?
Orako escalates: a nudge first, then an alert in your team channel. Nothing falls through.
How does the agent reuse a past answer?
Every conversation is saved as searchable history, scoped to your org's permissions. Before pinging a human, the agent searches it and reuses the matching answer, with who said it and when, so the same question is never asked twice.
Which tools does it work with?
Claude Code, Codex, and any MCP client connect over remote MCP: register the server and authorize once inside the agent, no token to paste. Teammates answer in Slack, Discord, or the dashboard.
Do you track live presence?
No. We don't guess who's around: whoever's there answers. Availability is proven by the reply, never by a status dot.
Put the human back in the loop
Give your agents the one thing they can’t generate: the answer that only your team knows.
No credit card · set up in minutes · or self-host it, free and open source