Refactor GatewayServer to serve HTTP and WebSocket on a shared
http.Server. Add static file serving with path traversal protection,
a dark-themed dashboard (system health, sessions, tools) and a
WebSocket chat interface with streaming tool events and markdown
rendering.
Phase 2 of the Flynn roadmap. Adds a WebSocket gateway server that
starts alongside the Telegram bot, providing real-time API access to
the agent, sessions, and tools.
Protocol: JSON-RPC-like (request/response/event) over WebSocket.
8 methods: agent.send, agent.cancel, sessions.list, sessions.history,
sessions.create, tools.list, tools.invoke, system.health.
Auth: Bearer token + Tailscale identity header support.
Session bridge: per-connection agent instances with shared model router.
New files: src/gateway/ (protocol, router, server, auth, session-bridge,
handlers for agent/sessions/tools/system).
57 new tests (181 total), typecheck clean.
Telegram bot now shows tool status during execution:
- Sends status message when tool starts (tool name + args snippet)
- Edits status message with result on completion
- Keeps typing indicator active during tool execution
- Adds setOnToolUse() to NativeAgent for per-message callback control
TUI now creates a NativeAgent with tool registry/executor and uses
agent.process() for message handling. Tool calls display status lines
showing tool name, args, and success/error result. Falls back to
direct model client when agent is not configured.
Lists available tools (shell, file ops, web fetch) so the model knows
what it can do. Keeps SOUL.md as the single source of truth for the
system prompt.
Tests the full agent -> tool execution -> model feedback loop:
- Shell tool execution with mock model
- Multi-tool chaining (file.write -> file.read) across iterations
- Verification that tool results are correctly passed back to model
Rewrites NativeAgent.process() from single-turn to an iterative tool
loop. When toolRegistry and toolExecutor are provided, the agent calls
the model, executes any requested tool calls, feeds results back, and
loops until the model returns a text response or max iterations hit.
- Backward compatible: works exactly as before without tools
- Supports onToolUse callback for frontend status display
- Max iterations (default 10) prevents infinite loops
- Handles multiple tool calls per model response
- 5 new tests (8 total)
Add LlamaCppClient to imports and handle llamacpp provider in createModelRouter.
Pass localProviders and currentLocalProvider configuration to MinimalTui to enable
the /backend command to access alternate provider configs.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add support for /backend command to show or switch local backend providers. Follows the same pattern as /model command with optional argument support.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Design for adding LlamaCppClient to support local LLM inference
via llama-server with CUDA. Target model: Qwen 2.5 14B Q4_K_M.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Streaming responses (chunk-based)
- Rich markdown rendering with syntax highlighting
- Model switching via /model command
- Scrollable message history
- Enhanced status bar with token usage
- Inline error handling with auto-retry
- Feature parity between minimal and fullscreen modes
William Valentin