Merge verified Odysseus fixes into dev

Dockerfile

@@ -16,7 +16,8 @@ FROM python:3.14-slim
 # downloads, and serves from Docker installs.
 # git/cmake are required when Cookbook builds llama.cpp on first llama.cpp
 # launch inside Docker.
-# nodejs/npm provide npx for the optional built-in Browser MCP server.
+# nodejs/npm provide npx for the built-in Browser MCP server.
+# chromium provides the actual browser binary used by that MCP server.
 # gosu lets the entrypoint drop privileges cleanly so signals still reach
 # uvicorn directly (no extra shell layer like `su`/`sudo` would add).
 RUN apt-get update && apt-get install -y --no-install-recommends \
@@ -26,6 +27,7 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     git \
     nodejs \
     npm \
+    chromium \
     tmux \
     openssh-client \
     gosu \

ROADMAP.md

@@ -32,6 +32,14 @@ the codebase, you are probably right to stay away.
   before the user request really starts. We need slimmer prompts, better tool
   selection, smaller default tool sets, and clearer guidance for models with
   4k/8k/16k context windows.
+- Local model speculative decoding support. For Odysseus-tuned local models,
+  plan to ship or recommend a small same-tokenizer draft model when the serving
+  backend supports it. Early vLLM testing showed a generic `Qwen3-0.6B` draft
+  beside `Qwen3-8B` can materially reduce wall time, while an unsupported
+  DSpark conversion performed poorly. Treat this as a supported draft-model lane
+  first; keep MTP-specific packaging as future work only when the architecture
+  and runtime support are real. Judge this by time-to-success, tool correctness,
+  grammar, and unchanged target output, not tokens/sec alone.
 - Skill/tool prompt-injection audit. User-editable skills, notes, documents,
   fetched pages, and memories should be treated as untrusted data. Keep testing
   whether models follow malicious instructions from those surfaces.

core/session_manager.py

@@ -543,6 +543,12 @@ def delete_session(self, session_id: str) -> bool:
         """Permanently delete a session and all its messages."""
         db = SessionLocal()
         try:
+            try:
+                from src.session_image_cleanup import cleanup_session_images
+                cleanup_session_images(session_id, db=db)
+            except Exception as e:
+                logger.warning(f"Image cleanup failed while deleting session {session_id}: {e}")
+
             # Detach documents so they survive as orphans in the library
             db.query(DbDocument).filter(DbDocument.session_id == session_id).update(
                 {DbDocument.session_id: None}, synchronize_session=False