# Roadmap 

<p align="center">
 <code>May 2026</code>
</p>

---

## In Progress

- [ ] `openmono stop` not working, it is not stopping llama-server correctly
- [ ] Colored diff display for file edits
- [ ] Git branch + status injected into system prompt
- [ ] Built-in playbooks — commit, review, explain, debug
- [ ] VS Code extension — quick-access agent from the editor

---

## Coming Next

- [ ] Parallel sub-agent management — spawn and coordinate multiple agents concurrently
- [ ] Playwright MCP — browser automation as a native tool
- [ ] MCP manager — discover, add, and configure MCP servers from within the agent
- [ ] Extended thinking + budget token support
- [ ] `/commit` and `/review` slash commands
- [ ] Background agent system with output streaming
- [ ] Micro-compaction before full compaction triggers

---

## On the Horizon

- [ ] Session revert — roll back conversation and file state to any prior turn
- [ ] Improved CLI — richer file read/write primitives, MultiEdit batch operations
- [ ] Theme system — light/dark, custom colour schemes
- [ ] Vim keybinding mode
- [ ] `/doctor` — diagnostics and connectivity check
- [ ] More providers — Google Vertex, Azure OpenAI, Bedrock, Groq
- [ ] Autocomplete with frecency scoring
- [ ] Session tagging and forking
- [ ] File watching during active sessions
- [ ] ACP (Agent Client Protocol) support for editor integrations such as Zed
- [ ] Desktop app — Tauri or Electron wrapper
- [ ] Web frontend for remote access
- [ ] Slack and GitHub App integrations
- [ ] Auto-dreaming — background memory consolidation
- [ ] Opt-in OpenTelemetry tracing

---

---

## Done

- [x] Agent iteration limit raised to 1000 and is now configurable — previously capped at 25 in [`ConversationLoop.cs`](https://github.com/StartupHakk/OpenMonoAgent.ai/blob/main/src/OpenMono.Cli/Session/ConversationLoop.cs)
- [x] Doom loop detection improved — aborts early if the same tool sequence repeats, preventing runaway agent loops
- [x] Plan mode integrated — restricts the agent to read-only tools for safe exploration before making changes
- [x] 12 GB and 16 GB GPU support — installer now detects VRAM tier and selects the appropriate model automatically
- [x] Vision / multimodal input — `@image.png` syntax in chat, `FileRead` for images, mmproj auto-downloaded at setup, smart resize to fit VRAM budget

---

*[Contributing](CONTRIBUTING.md)*