diff --git a/README.md b/README.md
index 2fdb6e9..f56f5a1 100644
--- a/README.md
+++ b/README.md
@@ -20,6 +20,7 @@ Self-hosted personal AI assistant with Telegram and Terminal interfaces.
 - **Capture Tools**: `screen.capture` and `camera.capture` tools for host capture workflows
 - **Session Transfer**: Move conversations between frontends
 - **CLI**: Full command-line interface (`flynn start`, `send`, `doctor`, `completion`, etc.)
+- **Optional Pi Embedded Backend**: Canary-only in-process Pi runtime path (`pi_embedded`) with native fallback
 - **Shell Completion**: Auto-generated completions for bash, zsh, and fish with `--install` flag
 - **Cron Scheduling**: Automated messages on cron schedules with output routing
 - **Daily Briefing Automation**: Optional built-in morning briefing preset (calendar + inbox + tasks summary prompt)
@@ -351,15 +352,22 @@ backends:
   claude_code: { enabled: false, path: /usr/local/bin/claude, args: [], timeout_ms: 120000 }
   opencode: { enabled: false, path: /usr/local/bin/opencode, args: [], timeout_ms: 120000 }
   gemini: { enabled: false, path: /usr/local/bin/gemini, args: [], timeout_ms: 120000 }
+  pi_embedded:
+    enabled: false
+    timeout_ms: 120000
+    no_tools_mode: true
+    model: openclaw-default
+    system_prompt_mode: hybrid   # flynn | pi_default | hybrid
+    module: "@badlogic/pi-agent-core"  # optional module override
 ```
 
-Each external backend also supports `retries` and `retry_delay_ms` for transient CLI failures.
+`pi_embedded` is intended for canary migration cohorts. In spike mode (`no_tools_mode: true`), Flynn keeps tool-oriented turns on native and only routes plain-text turns to Pi.
 
 When `args` is non-empty:
 - use `{prompt}` in an argument to inject the full generated prompt directly into argv.
-- if `{prompt}` is not present, Flynn writes the prompt to stdin.
+- if `{prompt}` is not present, Flynn appends backend-specific prompt args.
 
-If multiple external backends are enabled, set `backends.default` to choose explicitly. If omitted, Flynn selects by priority: `codex` -> `claude_code` -> `opencode` -> `gemini`.
+If multiple external backends are enabled, set `backends.default` to choose explicitly. If omitted, Flynn selects by priority: `codex` -> `claude_code` -> `opencode` -> `gemini` -> `pi_embedded`.
 
 You can also route specific named agents to a backend:
 
@@ -367,7 +375,10 @@ You can also route specific named agents to a backend:
 agent_configs:
   coder:
     model_tier: complex
-    backend: codex   # native | codex | claude_code | opencode | gemini
+    backend: codex   # native | codex | claude_code | opencode | gemini | pi_embedded
+  pi_canary:
+    model_tier: default
+    backend: pi_embedded
 ```
 
 ### Native Audio Support
diff --git a/docs/api/PROTOCOL.md b/docs/api/PROTOCOL.md
index e69e370..7a6da1e 100644
--- a/docs/api/PROTOCOL.md
+++ b/docs/api/PROTOCOL.md
@@ -36,6 +36,7 @@ The gateway serialises agent work **per session**, not per WebSocket connection:
 - Requests for different sessions can run in parallel.
 - Lane policy is configurable (`collect`, `followup`, `steer`, `steer_backlog`, `interrupt`) with per-channel and per-session overrides.
 - Session-local overrides can be managed at runtime via `agent.send` commands: `/queue`, `/queue set ...`, `/queue reset`.
+- Backend selection for a turn is server-side (`native` by default, optional external backends per config: `claude_code`, `opencode`, `codex`, `gemini`, `pi_embedded`) and does not change JSON-RPC method signatures.
 
 This is implemented via a per-lane queue (`LaneQueue`) in the gateway server, and used by `agent.send` and `agent.cancel`.
 
diff --git a/docs/architecture/AGENT_DIAGRAM.md b/docs/architecture/AGENT_DIAGRAM.md
index 1e4b2ce..31f437b 100644
--- a/docs/architecture/AGENT_DIAGRAM.md
+++ b/docs/architecture/AGENT_DIAGRAM.md
@@ -31,6 +31,7 @@ flowchart LR
     SM[SessionManager\nSQLite]
     OR[AgentOrchestrator]
     NA[NativeAgent\n(tool loop)]
+    EB[Optional External Backends\nclaude_code/opencode/codex/gemini/pi_embedded]
     MR[ModelRouter]
     TP[ToolPolicy + ToolRegistry]
     TE[ToolExecutor\nhooks + enforcement + audit]
@@ -60,7 +61,9 @@ flowchart LR
   CA --> RT
   RT --> SM
   RT --> OR
+  RT --> EB
   OR --> NA
+  EB --> MP
   NA --> MR
   MR --> MP
 
@@ -107,6 +110,9 @@ ChannelAdapter  ->  ChannelRegistry
         |               v
         |          ModelClient
         |
+        +----> (optional, non-tool turns) ExternalBackend
+                         (claude_code/opencode/codex/gemini/pi_embedded)
+        |
         +----> (optional) PairingManager gate for unknown senders
 
 Tool Calls (inside NativeAgent loop)
@@ -130,6 +136,7 @@ Key files:
 - Routing + per-session agent creation: `src/daemon/routing.ts`
 - Orchestration: `src/backends/native/orchestrator.ts`
 - Tool loop: `src/backends/native/agent.ts`
+- External backend adapters: `src/backends/external.ts`, `src/backends/piEmbedded.ts`
 - Model routing: `src/models/router.ts`
 - Tool policy + execution: `src/tools/policy.ts`, `src/tools/executor.ts`
 
diff --git a/docs/architecture/GATEWAY_SESSIONS_AND_QUEUE.md b/docs/architecture/GATEWAY_SESSIONS_AND_QUEUE.md
index d87c87c..874a92d 100644
--- a/docs/architecture/GATEWAY_SESSIONS_AND_QUEUE.md
+++ b/docs/architecture/GATEWAY_SESSIONS_AND_QUEUE.md
@@ -10,6 +10,7 @@ If you only want the protocol surface, see `docs/api/PROTOCOL.md`.
 - Each connection is attached to a `sessionId`.
 - Agent work is queued per `sessionId` (FIFO), not per connection.
 - Sessions persist in SQLite via `SessionManager` even if clients disconnect.
+- Once dequeued, message routing may execute the native orchestrator path or an optional external backend path (`claude_code`, `opencode`, `codex`, `gemini`, `pi_embedded`) depending on agent/backend config.
 
 ## Component Map
 
@@ -30,7 +31,7 @@ flowchart LR
   subgraph CORE[Flynn Core]
     SM[SessionManager\nin-memory cache + SQLite]
     SS[SessionStore\nSQLite tables]
-    AO[AgentOrchestrator]
+    AO[AgentOrchestrator / External Backends]
   end
 
   WS --> GS
diff --git a/docs/plans/state.json b/docs/plans/state.json
index cb17b72..1b97795 100644
--- a/docs/plans/state.json
+++ b/docs/plans/state.json
@@ -3,6 +3,33 @@
   "updated_at": "2026-02-24",
   "description": "Tracks the status of all Flynn plans and implementation phases",
   "plans": {
+    "pi-embedded-backend-canary-spike": {
+      "status": "completed",
+      "date": "2026-02-24",
+      "updated": "2026-02-24",
+      "summary": "Implemented a Pi embedded canary spike with a new optional `pi_embedded` backend, guarded no-tools canary routing, backend success/fallback latency telemetry in audit logs, focused backend/schema/routing tests, and architecture/protocol documentation updates while keeping native orchestration as the default path.",
+      "files_modified": [
+        "src/backends/piEmbedded.ts",
+        "src/backends/piEmbedded.test.ts",
+        "src/backends/external.ts",
+        "src/backends/index.ts",
+        "src/daemon/index.ts",
+        "src/daemon/routing.ts",
+        "src/daemon/routing.test.ts",
+        "src/config/schema.ts",
+        "src/config/schema.test.ts",
+        "src/agents/registry.ts",
+        "src/audit/types.ts",
+        "src/audit/logger.ts",
+        "config/default.yaml",
+        "README.md",
+        "docs/architecture/AGENT_DIAGRAM.md",
+        "docs/architecture/GATEWAY_SESSIONS_AND_QUEUE.md",
+        "docs/api/PROTOCOL.md",
+        "docs/plans/state.json"
+      ],
+      "test_status": "pnpm test:run src/backends/piEmbedded.test.ts src/config/schema.test.ts src/daemon/routing.test.ts + pnpm typecheck + pnpm lint (warnings only) passing"
+    },
     "full-audit-hardening-and-config-consolidation": {
       "status": "completed",
       "date": "2026-02-24",
@@ -6403,7 +6430,7 @@
     }
   },
   "overall_progress": {
-    "total_test_count": 1982,
+    "total_test_count": 1989,
     "all_tests_passing": true,
     "p0_completion": "3/3 (100%)",
     "p1_completion": "4/4 (100%)",
@@ -6433,7 +6460,8 @@
     "model_router_correctness": "completed — fallback paths now avoid duplicate clients, apply retry policy consistently, and reject unsupported OpenAI OAuth tool requests early",
     "native_audio_support": "completed — smart routing for native audio (Gemini/OpenAI/GitHub) vs Whisper transcription fallback, plus 2026-02-23 arg hydration hardening, tool.args_rewritten audit metric, transient fetch retry/timeout hardening, localhost->127.0.0.1 fallback for transcription endpoint connectivity, and whisper docker-compose entrypoint arg fix for port 18801",
     "remaining_phases_completion": "Phase 1: 3/3 (100%) — context levels, command registry, memory structure. Phase 2: 3/3 (100%) — component registry, confidence routing, history index. Phase 3: 2/2 (100%) — adaptive memory/compaction, truthfulness/autonomy hardening",
-    "next_up": "Track OpenClaw evolution regularly for inspiration and feature ideas"
+    "next_up": "Track OpenClaw evolution regularly for inspiration and feature ideas",
+    "pi_embedded_canary_spike": "completed — added optional pi_embedded backend adapter, canary-safe no-tools routing guard, backend success/fallback latency audit events, and docs/diagram updates while native remains default"
   },
   "soul_md_and_cron_create": {
     "date": "2026-02-11",