|
| 1 | +--- |
| 2 | +title: 'Introducing OpenAB: Your AI Coding Agent, Now in Your Team Chat' |
| 3 | +description: 'OpenAB is the chat-native runtime layer for ACP coding agents. An open-source broker that connects any ACP-compatible coding CLI to Discord, Slack, and other chat platforms.' |
| 4 | +pubDate: 'May 19 2026' |
| 5 | +--- |
| 6 | + |
| 7 | +*Your IDE is powerful. Your chat is where decisions happen. OpenAB connects the two.* |
| 8 | + |
| 9 | +--- |
| 10 | + |
| 11 | +## The problem |
| 12 | + |
| 13 | +AI coding agents are transforming how software gets built. Claude Code can refactor a module. Kiro CLI can scaffold an entire feature. Codex can fix a bug from a one-line description. |
| 14 | + |
| 15 | +But there's a gap: these agents live in terminals. Your team lives in Discord and Slack. |
| 16 | + |
| 17 | +When a junior developer hits a wall at 2 AM, they don't open a terminal — they post in the team channel. When a PM wants to understand why a deploy failed, they ask in the ops thread. When a code review needs a second opinion, the conversation happens in chat. |
| 18 | + |
| 19 | +Today, bridging that gap means copy-pasting between tools, losing context, and breaking flow. The agent can't see the conversation history. The team can't see what the agent did. Nobody wins. |
| 20 | + |
| 21 | +## What OpenAB does |
| 22 | + |
| 23 | +**OpenAB is the chat-native runtime layer for ACP coding agents.** It's an open-source broker that connects any [Agent Client Protocol (ACP)](https://github.com/agentclientprotocol/agent-client-protocol)-compatible coding CLI to Discord, Slack, and other chat platforms. |
| 24 | + |
| 25 | +ACP is a standard that lets AI coding tools communicate over a simple message protocol — think of it as a universal plug between chat apps and coding agents. |
| 26 | + |
| 27 | +``` |
| 28 | +┌──────────────┐ ┌──────────────┐ ┌──────────────────┐ |
| 29 | +│ Discord │◄────────────►│ │────────────►│ coding CLI │ |
| 30 | +│ Slack │ │ openab │◄───────────│ (acp mode) │ |
| 31 | +│ Telegram │◄──webhook───►│ (Rust) │ │ │ |
| 32 | +│ LINE │ │ │ │ kiro / claude / │ |
| 33 | +│ Feishu │ └──────────────┘ │ codex / gemini / │ |
| 34 | +│ Google Chat│ │ copilot / hermes │ |
| 35 | +└──────────────┘ └──────────────────┘ |
| 36 | +``` |
| 37 | + |
| 38 | +You `@mention` the bot in a channel. OpenAB spawns a thread, starts the agent, and streams responses back — live-edited every 1.5 seconds as tokens arrive. Follow-up messages in the thread go straight to the agent. No @mention needed. |
| 39 | + |
| 40 | +The agent gets the full thread context. The team sees progress and output in real time. Context stays intact. |
| 41 | + |
| 42 | +## Why this matters |
| 43 | + |
| 44 | +### Your team already uses chat |
| 45 | + |
| 46 | +Developers, PMs, designers — everyone's in Discord or Slack. Putting the agent where the team already is means: |
| 47 | + |
| 48 | +- **Zero onboarding friction.** No new tool to install. Type a message, get code. |
| 49 | +- **Shared context.** The whole team sees what the agent did and why. No more "what did you ask it?" |
| 50 | +- **Async-friendly.** Start a task before lunch. Come back to a completed PR in the thread. |
| 51 | + |
| 52 | +### One config, any agent |
| 53 | + |
| 54 | +OpenAB doesn't lock you into a single AI provider. Swap the backend by changing one line in `config.toml`: |
| 55 | + |
| 56 | +```toml |
| 57 | +[agent] |
| 58 | +command = "kiro-cli" # or "claude-agent-acp", "codex-acp", "gemini --acp" |
| 59 | +args = ["acp", "--trust-all-tools"] |
| 60 | +``` |
| 61 | + |
| 62 | +Supported agents today: |
| 63 | + |
| 64 | +| Agent | Command | Status | |
| 65 | +|-------|---------|--------| |
| 66 | +| Kiro CLI | `kiro-cli acp` | ✅ Verified | |
| 67 | +| Claude Code | `claude-agent-acp` | ✅ Verified | |
| 68 | +| Codex | `codex-acp` | ✅ Verified | |
| 69 | +| Gemini | `gemini --acp` | ✅ Verified | |
| 70 | +| Copilot CLI | `copilot --acp --stdio` | ✅ Verified | |
| 71 | +| Hermes | `hermes-acp` | ✅ Verified | |
| 72 | +| Grok Build | `grok agent stdio` | 🧪 Community | |
| 73 | +| OpenCode | `opencode acp` | 🧪 Community | |
| 74 | +| Cursor | `cursor-agent acp` | 🧪 Community | |
| 75 | + |
| 76 | +> *Status reflects integration testing as of publication. Community-contributed adapters may require additional setup.* |
| 77 | +
|
| 78 | +### Multi-agent collaboration |
| 79 | + |
| 80 | +Teams can run multiple agents in the same server. With the right configuration, they can coordinate work through threads: |
| 81 | + |
| 82 | +``` |
| 83 | +@JARVIS implement the login feature |
| 84 | + └─ JARVIS writes code, @mentions FRIDAY for review |
| 85 | + └─ FRIDAY reviews, flags an issue |
| 86 | + └─ JARVIS fixes, marks ready for merge |
| 87 | +``` |
| 88 | + |
| 89 | +Bot-to-bot messaging is built in. Each agent runs in its own process with its own session. The specifics depend on your agent and gateway setup — we'll cover multi-agent patterns in a dedicated post. |
| 90 | + |
| 91 | +## Architecture |
| 92 | + |
| 93 | +OpenAB is written in Rust. A small, self-contained binary with no language runtime dependency. |
| 94 | + |
| 95 | +Key design decisions: |
| 96 | + |
| 97 | +- **Session pool** — one CLI process per thread, auto-managed lifecycle. Configurable max sessions and TTL. |
| 98 | +- **Edit-streaming** — the Discord message updates live as tokens arrive, not after the full response completes. |
| 99 | +- **Emoji status** — 👀 (seen) → 🤔 (thinking) → 🔥 (coding) → 👍 (done). Your team knows what the agent is doing at a glance. |
| 100 | +- **Thread isolation** — each conversation is a separate thread with its own agent session. No cross-contamination. |
| 101 | + |
| 102 | +``` |
| 103 | +┌─ Kubernetes Pod ──────────────────────────────────────┐ |
| 104 | +│ openab (PID 1) │ |
| 105 | +│ └─ kiro-cli acp (child process) │ |
| 106 | +│ ├─ stdin ◄── JSON-RPC requests │ |
| 107 | +│ └─ stdout ──► JSON-RPC responses │ |
| 108 | +│ │ |
| 109 | +│ PVC (/data) │ |
| 110 | +│ ├─ ~/.kiro/ (settings, sessions) │ |
| 111 | +│ └─ ~/.local/share/kiro-cli/ (OAuth tokens) │ |
| 112 | +└───────────────────────────────────────────────────────┘ |
| 113 | +``` |
| 114 | + |
| 115 | +## Getting started |
| 116 | + |
| 117 | +### Prerequisites |
| 118 | + |
| 119 | +1. A Discord bot token (or Slack bot + app token) |
| 120 | +2. The coding CLI of your choice installed |
| 121 | +3. Docker (for local dev) or a Kubernetes cluster (for production) |
| 122 | + |
| 123 | +### Run locally with Docker (2 minutes) |
| 124 | + |
| 125 | +```bash |
| 126 | +docker run -e DISCORD_BOT_TOKEN=your-token \ |
| 127 | + -v $(pwd)/config.toml:/etc/openab/config.toml \ |
| 128 | + ghcr.io/openabdev/openab:latest |
| 129 | +``` |
| 130 | + |
| 131 | +### Deploy with Helm (production) |
| 132 | + |
| 133 | +```bash |
| 134 | +helm repo add openab https://openabdev.github.io/openab |
| 135 | +helm repo update |
| 136 | + |
| 137 | +helm install openab openab/openab \ |
| 138 | + --set agents.kiro.discord.botToken="$DISCORD_BOT_TOKEN" \ |
| 139 | + --set-string 'agents.kiro.discord.allowedChannels[0]=YOUR_CHANNEL_ID' |
| 140 | +``` |
| 141 | + |
| 142 | +### Authenticate the agent (first time only) |
| 143 | + |
| 144 | +```bash |
| 145 | +kubectl exec -it deployment/openab-kiro -- kiro-cli login --use-device-flow |
| 146 | +kubectl rollout restart deployment/openab-kiro |
| 147 | +``` |
| 148 | + |
| 149 | +### Use it |
| 150 | + |
| 151 | +In your Discord channel: |
| 152 | + |
| 153 | +``` |
| 154 | +@YourBot explain the auth middleware in src/auth.rs |
| 155 | +``` |
| 156 | + |
| 157 | +The bot creates a thread and starts streaming the response. That's it. |
| 158 | + |
| 159 | +## Beyond the basics |
| 160 | + |
| 161 | +OpenAB ships with more than just chat-to-agent bridging. Scheduled messages let you set up cron-driven prompts (daily standups, weekly PR digests) without an external scheduler. Voice messages are auto-transcribed so you can talk to your agent. Screenshots and files sent in chat are passed directly to the agent. And a custom gateway service extends support to Telegram, LINE, Feishu, and Google Chat. |
| 162 | + |
| 163 | +Each of these deserves its own deep-dive — we'll cover them in future posts. |
| 164 | + |
| 165 | +## The stack |
| 166 | + |
| 167 | +- **Language:** Rust |
| 168 | +- **Protocol:** Agent Client Protocol (ACP) |
| 169 | +- **Platforms:** Discord (Gateway WS), Slack (Socket Mode), Telegram/LINE/Feishu/Google Chat (Custom Gateway) |
| 170 | +- **Deployment:** Helm chart, raw K8s manifests, or Docker |
| 171 | +- **License:** MIT |
| 172 | + |
| 173 | +## What's next |
| 174 | + |
| 175 | +- Improved context management for long-running sessions |
| 176 | +- First-class MCP (Model Context Protocol) server integration per agent |
| 177 | +- Web dashboard for session monitoring |
| 178 | +- More platform gateways |
| 179 | + |
| 180 | +## Get involved |
| 181 | + |
| 182 | +OpenAB is 100% open-source under the MIT license. No CLA required. |
| 183 | + |
| 184 | +- **GitHub:** [github.com/openabdev/openab](https://github.com/openabdev/openab) |
| 185 | +- **Discord:** [Join the community](https://discord.gg/DmbhfDZjQS) |
| 186 | +- **Contributing:** [CONTRIBUTING.md](https://github.com/openabdev/openab/blob/main/CONTRIBUTING.md) |
| 187 | + |
| 188 | +We're especially looking for feedback from teams already running AI agents in their workflow. What's missing? What would make this indispensable instead of just interesting? |
| 189 | + |
| 190 | +Come build with us. |
0 commit comments