README: expand remote execution section with streaming, persistence, comparison

README.md

@@ -149,11 +149,20 @@ Monorepo with Git submodules. Each submodule has its own Go module (except `el`
 ### Remote execution (split-brain)
 
 - **Local brain, remote hands** — the agent loop and LLM streaming stay local; tool execution happens on a remote host over SSH
-- **Zero-setup deployment** — `ollie-remote` is bootstrapped automatically on first connect (binary cached by hash)
+- **Zero-setup deployment** — `ollie-remote` is bootstrapped automatically on first connect (binary + tools cached by hash)
+- **Streaming output** — real-time command output over the SSH channel; builds and tests display as they run
+- **Session persistence** — remote sessions survive server restarts; reconnect automatically on restore
 - **Sandboxed remotely** — Landlock enforcement runs on the remote host; `landrun` is shipped alongside the binary
 - **Convention over configuration** — sensible sandbox defaults; per-project overrides via `.ollie-sandbox.yaml` in the project root
 - **GUI works unchanged** — D-Bus signals fire locally, so KDE GUI / web UI see real-time progress
 
+The split-brain model complements 9P's network transparency. 9P mounts a remote agent as a local filesystem (remote brain, local view); split-brain keeps the brain local and sends execution remote (local brain, remote hands). Use 9P for monitoring fleets of headless agents; use split-brain for interactive development on remote hosts.
+
+```
+9P remote mount:       you ──9P──▶ remote olliesrv ──▶ remote tools
+Split-brain:           local olliesrv ──SSH──▶ remote ollie-remote ──▶ remote tools
+```
+
 See [doc/REMOTE_EXECUTION.md](doc/REMOTE_EXECUTION.md) for the design.
 
 ### Stateless endpoints