OpenRyoko is an open-source (MIT-licensed) daemon for AI agents published by TEKION Group Inc. It places coding agents such as Claude Code, Codex, and Gemini CLI in Slack as an “AI coworker” that stays out of casual chat and acts only when called. It is designed to fix three common frustrations of using AI inside conversations: it interrupts too much, it stops midway, and you cannot see its progress. It was released in April 2026, and according to the GitHub release notes, updates such as support for Claude Opus 5.5 continued into September 2026.
Key Features
- Read-the-room triage: For each Slack message, a lightweight model decides whether to stay silent (silent), just add a reaction (react), or reply (reply). The design is conservative: it does not respond when its confidence is below 60%, so it does not chime in too often even in casual channels
- Natural-language /goal: From a request like “see this through to the end,” it extracts completion criteria and keeps the session going until they are met. It uses Claude Code’s /goal and Codex’s completion checks to keep work from stalling halfway
- Progress in Slack Canvas: The “Agents View Canvas” syncs the list of running and waiting sessions to a Slack Canvas every 30 seconds. You can check progress inside Slack without opening a separate dashboard
- Engine switching: You can switch between Claude Code CLI, Codex, and Gemini CLI from the settings screen, choosing based on the nature of the task and cost
- AI organizations, Skills, Cron, and MCP connections: It supports “AI organizations” with departments and ranks, adding skills in Markdown, scheduled tasks via cron, and connections to external MCP servers such as GitHub and Notion (OpenRyoko connects as an MCP client rather than acting as an MCP server itself)
Pricing
OpenRyoko itself is free, open-source software. Separately, the developer officially sells one-time-purchase “packages” of operational know-how. The packages are not the software itself but sets of configuration assets such as operating instructions, persona files, and Skills, delivered via an invitation to a private GitHub repository.
| Plan | Price | Main contents |
|---|---|---|
| OpenRyoko core | Free | All features, self-hosted |
| Basic package | ¥29,800 (one-time, tax included) | Operating instructions, 5 persona files, cron pattern collection, verification gates, 9 Skills, and more (26 files total) |
| Extended package | ¥98,000 (one-time, tax included) | Everything in the basic package plus a knowledge graph, an autonomous-loop kit, a Docker sandbox, Advanced Skills, and more |
Even though the core is free, the AI engines that actually run the work cost extra (Claude or OpenAI subscriptions, API usage fees).
Pricing is as of September 2026. For the latest pricing, please check the official website.
Pros & Cons
✅ Pros
- You can put an AI coworker in Slack on your own infrastructure without paying license fees
- Because engines are switchable, you are less exposed to price increases or spec changes from a single AI vendor
- It stays quiet even when resident in casual channels, so it is unlikely to disrupt team conversations
- You can track long-running work inside Slack, which saves you from asking “how far along is it?”
⚠️ Cons
- Transport encryption is not built in. If you expose it externally, you need to set up Tailscale, a VPN, or a reverse proxy yourself
- If you neglect the allowed-user settings (such as allowFrom), someone in your Slack workspace could abuse the bot’s permissions through prompt injection
- It is intended for an individual or a single team, not as the backend for a SaaS serving multiple organizations
- Running it requires Node.js 22+, pnpm, the Claude Code CLI, and so on, which makes setup somewhat difficult for non-engineers alone
- The operational know-how for production use is bundled in the paid packages, so reaching the same level of operation with the core alone requires your own design work
Comparison with Similar Services
The following is a simple comparison based on each service’s public information.
| Service | Difference from OpenRyoko |
|---|---|
| Jinn | The multi-CLI daemon foundation that OpenRyoko was forked from. The Slack-specific read-the-room behavior and /goal completion are OpenRyoko’s own additions |
| OpenClaw | A personal AI assistant platform supporting multiple channels such as WhatsApp, Telegram, and Discord. OpenRyoko goes further in refining behavior inside Slack |
| Hermes Agent | An agent centered on persistent memory and an autonomous learning loop, not designed around living in Slack |
Who Is It For
- Developers who already use Claude Code or Codex and want teammates to request work casually from Slack
- Teams that want to automate routine work such as PR status reports, competitor research, and monitoring with cron
- Indie developers who want to hand longer tasks to AI while watching progress in Slack
- People who can manage their own servers and take responsibility for security settings
Summary
OpenRyoko’s strength is that it directly addresses the frustrations of using AI in Slack: chiming in when nobody asked, and stopping partway through a requested task. Since the core is free to try, a good first step is to run it in a channel of your own and see how well it reads the room and how persistently /goal keeps going. When you move to serious team use, locking down access control and protecting the communication path first is essential.