diff --git a/docs/PLAN-INTERFACE.md b/docs/PLAN-INTERFACE.md
index 1493ebc..f9d774b 100644
--- a/docs/PLAN-INTERFACE.md
+++ b/docs/PLAN-INTERFACE.md
@@ -720,3 +720,54 @@ Earlier notes in this file said "software is not installed". That was wrong: Mou
 **hosted API**, not a program. What exists here is `MOUSE_API_KEY`; what is missing is a client
 integration — an HTTP call like any other. No install is needed, and nothing has to run locally.
 The old project's policy still applies: triage only, never a gate, never a permit.
+
+### Headroom: what the person has to do
+
+**Path A — the tool. Nothing to configure.** `headroom-ai` is installed for Pi on `.13`, `.27` and
+`.51`. To give it to one agent:
+
+1. `Agents>agents` → `)` to the **extensions** tab.
+2. `↑`/`↓` to `headroom-bridge`, press `space`. That writes a wrapper for that agent and sets its
+   command; no other agent changes.
+3. Run one task with that agent, then read the run log in `Agents>agents`: the tool call
+   `compress_for_agent` appears with the token numbers before and after.
+
+**Path B — the provider. One file in the headroom container.** Headroom is started from
+`~/Docker/Containers/headroom/docker-compose.yml` with no config mount. Add one:
+
+```yaml
+services:
+  headroom:
+    image: ghcr.io/chopratejas/headroom:latest
+    container_name: headroom
+    restart: always
+    command: --port 8787 --no-telemetry
+    ports: ['8787:8787']
+    volumes:
+      - ./headroom-config.yaml:/app/config.yaml:ro
+    environment:
+      - LITELLM_CONFIG=/app/config.yaml
+```
+
+and a `headroom-config.yaml` beside it that teaches LiteLLM the router's models:
+
+```yaml
+model_list:
+  - model_name: deepseek/deepseek-v4-flash
+    litellm_params:
+      model: openai/deepseek-v4-flash
+      api_base: http://192.168.20.13:20129/v1
+      api_key: os.environ/OMNI_KEY
+  - model_name: auto/best-fast
+    litellm_params:
+      model: openai/auto/best-fast
+      api_base: http://192.168.20.13:20129/v1
+      api_key: os.environ/OMNI_KEY
+```
+
+then restart the container. After that, add a provider in `~/.pi/agent/settings.json` pointing at
+`http://192.168.20.13:8787/v1`, and set one agent's model to `headroom/<model>`. That is the path
+the earlier test was trying, and it fails today only because LiteLLM does not know those ids.
+
+**What the person does:** choose A or B. A needs no file edits. B needs the two files above and a
+container restart, and then one agent's model changed.