feat(pi): support Pi coding agent via Dockerfile.pi

.github/workflows/build-operator.yml

@@ -75,6 +75,7 @@ jobs:
           - { suffix: "-hermes", dockerfile: "Dockerfile.hermes", artifact: "hermes" }
           - { suffix: "-grok", dockerfile: "Dockerfile.grok", artifact: "grok" }
           - { suffix: "-antigravity", dockerfile: "Dockerfile.antigravity", artifact: "antigravity" }
+          - { suffix: "-pi", dockerfile: "Dockerfile.pi", artifact: "pi" }
         platform:
           - { os: linux/amd64, runner: ubuntu-latest }
           - { os: linux/arm64, runner: ubuntu-24.04-arm }
@@ -141,6 +142,7 @@ jobs:
           - { suffix: "-hermes", artifact: "hermes" }
           - { suffix: "-grok", artifact: "grok" }
           - { suffix: "-antigravity", artifact: "antigravity" }
+          - { suffix: "-pi", artifact: "pi" }
     runs-on: ubuntu-latest
     permissions:
       contents: read
@@ -195,6 +197,7 @@ jobs:
           - { suffix: "-hermes" }
           - { suffix: "-grok" }
           - { suffix: "-antigravity" }
+          - { suffix: "-pi" }
     runs-on: ubuntu-latest
     permissions:
       contents: read

.github/workflows/docker-smoke-test.yml

@@ -23,6 +23,7 @@ jobs:
           - { dockerfile: Dockerfile.hermes, suffix: "-hermes", agent: "hermes-acp", agent_args: "" }
           - { dockerfile: Dockerfile.grok, suffix: "-grok", agent: "grok", agent_args: "agent stdio" }
           - { dockerfile: Dockerfile.antigravity, suffix: "-antigravity", agent: "agy-acp", agent_args: "" }
+          - { dockerfile: Dockerfile.pi, suffix: "-pi", agent: "pi-acp", agent_args: "" }
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v6

AGENTS.md

@@ -75,7 +75,7 @@ Never leak `DISCORD_BOT_TOKEN` or other OAB credentials to the agent.
 
 ### 4. Dockerfile Discipline
 
-There are 7 Dockerfiles: `Dockerfile`, `Dockerfile.claude`, `Dockerfile.codex`, `Dockerfile.copilot`, `Dockerfile.cursor`, `Dockerfile.gemini`, `Dockerfile.opencode`.
+There are 11 Dockerfiles: `Dockerfile`, `Dockerfile.antigravity`, `Dockerfile.claude`, `Dockerfile.codex`, `Dockerfile.copilot`, `Dockerfile.cursor`, `Dockerfile.gemini`, `Dockerfile.grok`, `Dockerfile.hermes`, `Dockerfile.opencode`, `Dockerfile.pi`.
 
 A change to one MUST be evaluated against ALL. Common layers (base image, openab binary, tini) are shared — update all or explain why not.
 

Dockerfile.pi

@@ -0,0 +1,38 @@
+# --- Build stage ---
+FROM rust:1-bookworm AS builder
+WORKDIR /build
+COPY Cargo.toml Cargo.lock ./
+RUN mkdir src && echo 'fn main() {}' > src/main.rs && cargo build --release && rm -rf src
+COPY src/ src/
+RUN touch src/main.rs && cargo build --release
+
+# --- Runtime stage ---
+FROM node:22-bookworm-slim
+RUN apt-get update && apt-get install -y --no-install-recommends ca-certificates curl git procps ripgrep tini bubblewrap socat && rm -rf /var/lib/apt/lists/*
+
+# Install pi-acp adapter and Pi coding agent CLI.
+# The Pi coding agent npm package was renamed from @mariozechner/pi-coding-agent to @earendil-works/pi-coding-agent.
+ARG PI_ACP_VERSION=0.0.27
+ARG PI_CODING_AGENT_VERSION=0.75.4
+RUN npm install -g pi-acp@${PI_ACP_VERSION} @earendil-works/pi-coding-agent@${PI_CODING_AGENT_VERSION} --retry 3
+
+# Install gh CLI (matches other Dockerfiles)
+RUN curl -fsSL https://cli.github.com/packages/githubcli-archive-keyring.gpg \
+      -o /usr/share/keyrings/githubcli-archive-keyring.gpg && \
+    echo "deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/githubcli-archive-keyring.gpg] https://cli.github.com/packages stable main" \
+      > /etc/apt/sources.list.d/github-cli.list && \
+    apt-get update && apt-get install -y --no-install-recommends gh && \
+    rm -rf /var/lib/apt/lists/*
+
+ENV HOME=/home/node
+WORKDIR /home/node
+
+COPY --from=builder --chown=node:node /build/target/release/openab /usr/local/bin/openab
+
+RUN mkdir -p /home/node/.pi && chown -R node:node /home/node/.pi
+
+USER node
+HEALTHCHECK --interval=30s --timeout=5s --retries=3 \
+  CMD pgrep -x openab || exit 1
+ENTRYPOINT ["tini", "--"]
+CMD ["openab", "run", "-c", "/etc/openab/config.toml"]

README.md

@@ -4,7 +4,7 @@
 
 ![OpenAB banner](images/banner.jpg)
 
-A lightweight, secure, cloud-native ACP harness that bridges **Discord, Slack**, and any [Agent Client Protocol](https://github.com/anthropics/agent-protocol)-compatible coding CLI (Kiro CLI, Claude Code, Codex, Gemini, OpenCode, Copilot CLI, Hermes, Grok Build, Antigravity, etc.) over stdio JSON-RPC — delivering the next-generation development experience. **Telegram, LINE, Feishu/Lark, Google Chat**, and other webhook-based platforms are supported via the standalone [Custom Gateway](gateway/).
+A lightweight, secure, cloud-native ACP harness that bridges **Discord, Slack**, and any [Agent Client Protocol](https://github.com/anthropics/agent-protocol)-compatible coding CLI (Kiro CLI, Claude Code, Codex, Gemini, OpenCode, Copilot CLI, Hermes, Grok Build, Antigravity, Pi, etc.) over stdio JSON-RPC — delivering the next-generation development experience. **Telegram, LINE, Feishu/Lark, Google Chat**, and other webhook-based platforms are supported via the standalone [Custom Gateway](gateway/).
 
 🪼 **Join our community!** Come say hi on Discord — we'd love to have you: **[🪼 OpenAB — Official](https://discord.gg/DmbhfDZjQS)** 🎉
 
@@ -22,8 +22,8 @@ A lightweight, secure, cloud-native ACP harness that bridges **Discord, Slack**,
 │   LINE       │◄──webhook──┌──────────────────┐              │ opencode acp     │
 │   User       │            │  Custom Gateway  │              │ grok agent stdio │
 ├──────────────┤            │  (standalone)    │              │ agy-acp          │
-│  Feishu/Lark │◄───WS──────│                  │              └──────────────────┘
-│   User       │            │                  │
+│  Feishu/Lark │◄───WS──────│                  │              │ pi-acp           │
+│   User       │            │                  │              └──────────────────┘
 ├──────────────┤            │                  │
 │ Google Chat  │◄──webhook──│                  │
 │   User       │            └──────────────────┘
@@ -38,7 +38,7 @@ A lightweight, secure, cloud-native ACP harness that bridges **Discord, Slack**,
 
 - **Multi-platform** — supports Discord and Slack, run one or both simultaneously
 - **Custom Gateway** — extend to Telegram, LINE, Feishu/Lark, Google Chat, MS Teams via standalone [gateway](gateway/)
-- **Pluggable agent backend** — swap between Kiro CLI, Claude Code, Codex, Gemini, OpenCode, Copilot CLI, Hermes, Grok Build, Antigravity via config
+- **Pluggable agent backend** — swap between Kiro CLI, Claude Code, Codex, Gemini, OpenCode, Copilot CLI, Hermes, Grok Build, Antigravity, Pi via config
 - **@mention trigger** — mention the bot in an allowed channel to start a conversation
 - **Thread-based multi-turn** — auto-creates threads; no @mention needed for follow-ups
 - **Multi-agent collaboration** — bot-to-bot messaging for coordinated workflows ([docs/multi-agent.md](docs/multi-agent.md))
@@ -171,6 +171,7 @@ The bot creates a thread. After that, just type in the thread — no @mention ne
 | Hermes Agent | `hermes-acp` | Native | [docs/hermes.md](docs/hermes.md) |
 | Grok Build | `grok agent stdio` | Native | [docs/grok.md](docs/grok.md) |
 | Antigravity | `agy-acp` | [agy-acp](agy-acp/) | [docs/antigravity.md](docs/antigravity.md) |
+| Pi | `pi-acp` | [pi-acp](https://www.npmjs.com/package/pi-acp) | [docs/pi.md](docs/pi.md) |
 
 > 🔧 Running multiple agents? See [docs/multi-agent.md](docs/multi-agent.md)
 

charts/openab/templates/NOTES.txt

@@ -39,6 +39,10 @@ Agents deployed:
 {{- else if eq (toString $cfg.command) "opencode" }}
     Authenticate:
     kubectl exec -it deployment/{{ include "openab.agentFullname" (dict "ctx" $ "agent" $name) }} -- opencode auth login
+{{- else if eq (toString $cfg.command) "pi-acp" }}
+    Authenticate:
+    kubectl exec -it deployment/{{ include "openab.agentFullname" (dict "ctx" $ "agent" $name) }} -- pi
+    (Once inside the interactive interface, type `/login` to authenticate)
 {{- else if eq (toString $cfg.command) "cursor-agent" }}
     Authenticate:
     kubectl exec -it deployment/{{ include "openab.agentFullname" (dict "ctx" $ "agent" $name) }} -- cursor-agent login

charts/openab/values.yaml

@@ -126,6 +126,28 @@ agents:
     #   agentsMd: ""
     #   resources: {}
     #   image: "ghcr.io/openabdev/openab-opencode:latest"
+    # pi:
+    #   command: pi-acp
+    #   discord:
+    #     enabled: true
+    #     allowedChannels:
+    #       - "YOUR_CHANNEL_ID"
+    #     allowedUsers: []
+    #   workingDir: /home/node
+    #   env: {}
+    #   envFrom: []
+    #   secretEnv: []
+    #   pool:
+    #     maxSessions: 10
+    #     sessionTtlHours: 24
+    #   reactions:
+    #     enabled: true
+    #     removeAfterReply: false
+    #   persistence:
+    #     enabled: true
+    #     storageClass: ""
+    #     size: 1Gi
+    #   image: "ghcr.io/openabdev/openab-pi:latest"
     # cursor:
     #   command: cursor-agent
     #   args:

docs/config-reference.md

@@ -95,7 +95,7 @@ The AI agent subprocess that OpenAB spawns to handle messages via ACP.
 
 | Key | Type | Default | Description |
 |-----|------|---------|-------------|
-| `command` | string | *required* | Agent binary (e.g. `kiro-cli`, `claude-agent-acp`, `codex`, `gemini`, `copilot`, `opencode`, `cursor-agent`). |
+| `command` | string | *required* | Agent binary (e.g. `kiro-cli`, `claude-agent-acp`, `codex`, `gemini`, `copilot`, `opencode`, `pi-acp`, `cursor-agent`). |
 | `args` | string[] | `[]` | CLI arguments passed to the agent. |
 | `working_dir` | string | `"/tmp"` | Working directory for the agent process. |
 | `env` | map | `{}` | Extra environment variables (e.g. `{ OPENAI_API_KEY = "${OPENAI_API_KEY}" }`). |
@@ -146,6 +146,11 @@ command = "opencode"
 args = ["acp"]
 working_dir = "/home/node"
 
+# Pi Agent
+[agent]
+command = "pi-acp"
+working_dir = "/home/node"
+
 # Cursor Agent
 [agent]
 command = "cursor-agent"

docs/pi.md

@@ -0,0 +1,109 @@
+# Pi Coding Agent
+
+OpenAB supports the [Pi coding agent](https://github.com/earendil-works/pi-coding-agent) via the `pi-acp` adapter — a Node.js bridge that translates ACP JSON-RPC into Pi CLI invocations.
+
+## Advantages Over Other Native Coding Agents
+
+Pi is a native coding agent that supports subscription-based authentication (like Codex, Cloud Code, and GitHub Copilot). Key advantages:
+
+### No Auth Proxy Required
+
+Pi natively supports Anthropic (Claude Pro/Max) and ChatGPT Plus/Pro subscriptions via OAuth. Unlike agents that require an `openab-auth-proxy` sidecar for subscription forwarding, Pi handles subscription auth directly — reducing deployment complexity and eliminating a moving part.
+
+| Agent | Subscription Auth | Auth Proxy Needed? |
+|-------|------------------|--------------------|
+| Pi | Native OAuth (`pi /login`) | ❌ No |
+| Codex | Native device flow | ❌ No |
+| GitHub Copilot | Native device flow | ❌ No |
+| Claude Code | Native OAuth | ❌ No |
+| Kiro | Native OAuth | ❌ No |
+
+### Minimal Tool Surface (Maximum Context Window)
+
+Pi exposes only 4 core tools: `read`, `write`, `edit`, `bash`. Combined with a tiny system prompt, this drastically reduces prompt overhead and maximizes the available context window for actual project source files.
+
+| Agent | Tool Count | System Prompt Size |
+|-------|-----------|-------------------|
+| Pi | 4 | Minimal |
+| Claude Code | 10+ | Large |
+| Codex | 8+ | Medium |
+| Copilot | 10+ | Large |
+
+### Multi-Model Support
+
+Pi is model-agnostic and supports 15+ LLM providers. Developers can switch models mid-session without restarting the agent or changing configuration.
+
+Supported providers include:
+- Anthropic (Claude) — via subscription or API key
+- OpenAI (GPT/Codex) — via subscription or API key
+- Google (Gemini) — via API key
+- Any OpenAI-compatible endpoint
+
+### Branching Session Trees
+
+Pi saves session history as trees, enabling clean branching of code exploration. This allows developers to explore multiple approaches from a single decision point without losing context.
+
+## Configuration
+
+```toml
+[agent]
+command = "pi-acp"
+working_dir = "/home/node"
+```
+
+## Docker
+
+```bash
+docker build -f Dockerfile.pi -t openab-pi:latest .
+```
+
+## Helm
+
+```yaml
+agents:
+  pi:
+    discord:
+      enabled: true
+      allowedChannels:
+        - "YOUR_CHANNEL_ID"
+    command: pi-acp
+    workingDir: /home/node
+    image: "ghcr.io/openabdev/openab-pi:latest"
+```
+
+## Authentication
+
+```bash
+kubectl exec -it deployment/openab-pi -- pi
+# Once inside the interactive interface, type /login to authenticate
+```
+
+Supported authentication methods:
+
+| Provider | Auth Method | Subscription |
+|----------|-------------|-------------|
+| Anthropic (Claude Pro/Max) | OAuth via `pi /login` | Claude subscription |
+| ChatGPT Plus/Pro | OAuth via `pi /login` | ChatGPT subscription |
+| Any API key provider | `env = { OPENAI_API_KEY = "..." }` | Pay-per-token |
+
+## Steering Files
+
+Pi reads steering files in this order:
+
+1. `.pi/SYSTEM.md` — replaces the default system prompt entirely
+2. `.pi/APPEND_SYSTEM.md` — appends to the default system prompt
+3. `AGENTS.md` — loaded hierarchically (project root → global) for context injection
+
+Place your steering instructions in `/home/node/AGENTS.md` or `/home/node/.pi/APPEND_SYSTEM.md`.
+
+## Persisted Paths (PVC)
+
+| Path | Contents |
+|------|----------|
+| `/home/node/.pi/` | Pi configuration and auth tokens |
+| `/home/node/.pi/sessions/` | Session history trees |
+
+## Limitations
+
+- **No streaming**: `pi-acp` returns the full response at once; streamed output is sent as a single `agent_message_chunk` notification.
+- **Cancel is best-effort**: Pi CLI runs to completion; `session/cancel` may not interrupt mid-generation.

docs/steering-design-guide.md

@@ -10,13 +10,13 @@ AI coding agents load persistent instructions every session, but without deliber
 
 This guide establishes a universal framework for organizing agent memory into layers, so rules are reliably followed, context budgets are respected, and teams can onboard new agents without starting from scratch.
 
-OpenAB is designed to be agent-agnostic — it supports Kiro, Claude Code, Codex, Gemini, Copilot, and OpenCode running side by side. This guide provides a shared memory architecture standard that allows all supported coding agents to maintain consistent behavior, collaborate effectively, and operate from a single source of truth regardless of their underlying platform differences.
+OpenAB is designed to be agent-agnostic — it supports Kiro, Claude Code, Codex, Gemini, Copilot, OpenCode, and Pi running side by side. This guide provides a shared memory architecture standard that allows all supported coding agents to maintain consistent behavior, collaborate effectively, and operate from a single source of truth regardless of their underlying platform differences.
 
 ---
 
 How to organize AI agent memory across three tiers: hot (always loaded), warm (triggered on demand), and cold (searched when needed).
 
-Applies to: Kiro, Claude Code, Codex, Gemini, Copilot, OpenCode — any agent that supports persistent instruction files.
+Applies to: Kiro, Claude Code, Codex, Gemini, Copilot, OpenCode, Pi — any agent that supports persistent instruction files.
 
 ---
 
@@ -165,6 +165,7 @@ Applies to: Kiro, Claude Code, Codex, Gemini, Copilot, OpenCode — any agent th
 | Gemini | `GEMINI.md` hierarchical (`~/.gemini/GEMINI.md` global → `./GEMINI.md` project → subdir) + `MEMORY.md` index | Same hierarchical pattern as CC/Codex. Private project memory index is hot; individual memory files are cold |
 | Copilot | `.github/copilot-instructions.md` (repo-wide) + `.github/instructions/**/*.instructions.md` (path-specific) + `AGENTS.md` (nearest-in-tree, where supported: cloud agent / CLI) | Layered: Personal > Path-specific > Repo-wide > Agent > Organization. No documented hard size cap for Chat/Agent (code review reads first 4K chars only). Keep short (~2 pages recommended) |
 | OpenCode | `AGENTS.md` or equivalent | Follows repo convention |
+| Pi | `AGENTS.md` hierarchical (project root → global) + `SYSTEM.md` or `APPEND_SYSTEM.md` in `.pi/` | Project or global `SYSTEM.md` replaces the default system prompt, while `APPEND_SYSTEM.md` appends to it. `AGENTS.md` is loaded hierarchically for context injection |
 
 ---