Roadmap

Implemented

The core AI agent loop and the payment-gateway layer are both shipped:

Four-stage pipeline — Translator → Planner → Executor → Summarizer
47 MCP tools — full Bitcoin Core + Core Lightning control surface
Multi-turn history — follow-up prompts resolve without restating context
Goal verification — automatic read-only confirmation after state-changing actions
Web UI — real-time dashboard with Pipeline tab, Network graph, Logs, Settings
SSE streaming — live pipeline events and LLM token stream
Cross-machine Lightning connectivity — sys_netinfo + ln_node_start with bind_host/announce_host
Four LLM backends — OpenAI, Anthropic Claude, Google Gemini, and local Ollama
x402 payment-gated endpoints — HTTP middleware issues BOLT11 invoices on 402 responses; the pipeline executor automatically pays (with a human-approval threshold) and retries.
Crash Kit — one-click debug snapshot from the web UI
Persisted smoke-test battery — 25 scripted prompts covering 42 of 47 tools.

Battery Phase 3 — LLM-planned multi-step payment flows
AI-coverage audit — tooling that verifies when the deterministic Planner dispatches can be demoted to fallback on stronger LLMs

Metrics — per-stage latency, payment success rate, tool-failure rate, LLM token counts
Security review — authentication, replay protection, and abuse prevention for payment-gated APIs beyond the current session-auth + CSRF + rate-limit stack
Multi-machine harness — scripted setup for running nodes on separate machines with cross-machine peer connectivity