feat: add multimodal image marker support with Ollama vision

docs/channels-reference.md

@@ -51,6 +51,22 @@ Notes:
 - Model cache previews come from `zeroclaw models refresh --provider <ID>`.
 - These are runtime chat commands, not CLI subcommands.
 
+## Inbound Image Marker Protocol
+
+ZeroClaw supports multimodal input through inline message markers:
+
+- Syntax: ``[IMAGE:<source>]``
+- `<source>` can be:
+  - Local file path
+  - Data URI (`data:image/...;base64,...`)
+  - Remote URL only when `[multimodal].allow_remote_fetch = true`
+
+Operational notes:
+
+- Marker parsing applies to user-role messages before provider calls.
+- Provider capability is enforced at runtime: if the selected provider does not support vision, the request fails with a structured capability error (`capability=vision`).
+- Linq webhook `media` parts with `image/*` MIME type are automatically converted to this marker format.
+
 ## Channel Matrix
 
 ---
@@ -349,4 +365,3 @@ If a specific channel task crashes or exits, the channel supervisor in `channels
 - `Channel message worker crashed:`
 
 These messages indicate automatic restart behavior is active, and you should inspect preceding logs for root cause.
-

docs/config-reference.md

@@ -62,6 +62,24 @@ Notes:
 - `reasoning_enabled = true` explicitly requests reasoning for supported providers (`think: true` on `ollama`).
 - Unset keeps provider defaults.
 
+## `[multimodal]`
+
+| Key | Default | Purpose |
+|---|---|---|
+| `max_images` | `4` | Maximum image markers accepted per request |
+| `max_image_size_mb` | `5` | Per-image size limit before base64 encoding |
+| `allow_remote_fetch` | `false` | Allow fetching `http(s)` image URLs from markers |
+
+Notes:
+
+- Runtime accepts image markers in user messages with syntax: ``[IMAGE:<source>]``.
+- Supported sources:
+  - Local file path (for example ``[IMAGE:/tmp/screenshot.png]``)
+- Data URI (for example ``[IMAGE:data:image/png;base64,...]``)
+- Remote URL only when `allow_remote_fetch = true`
+- Allowed MIME types: `image/png`, `image/jpeg`, `image/webp`, `image/gif`, `image/bmp`.
+- When the active provider does not support vision, requests fail with a structured capability error (`capability=vision`) instead of silently dropping images.
+
 ## `[gateway]`
 
 | Key | Default | Purpose |

docs/providers-reference.md

@@ -56,6 +56,13 @@ credential is not reused for fallback providers.
 | `lmstudio` | `lm-studio` | Yes | (optional; local by default) |
 | `nvidia` | `nvidia-nim`, `build.nvidia.com` | No | `NVIDIA_API_KEY` |
 
+### Ollama Vision Notes
+
+- Provider ID: `ollama`
+- Vision input is supported through user message image markers: ``[IMAGE:<source>]``.
+- After multimodal normalization, ZeroClaw sends image payloads through Ollama's native `messages[].images` field.
+- If a non-vision provider is selected, ZeroClaw returns a structured capability error instead of silently ignoring images.
+
 ### Bedrock Notes
 
 - Provider ID: `bedrock` (alias: `aws-bedrock`)