neuraloop 0.1.0

The only AI memory that gets smarter every night — SERA self-improvement loop, local-first, MCP-native

Neuraloop — Sovereign Developer Memory

The only AI memory that gets smarter every night. Local-first. MCP-native. Zero cloud. Your context stays yours.

---

What It Does

• Captures every coding session automatically — IDE stop hook, zero effort
• Distills your codebase history nightly via SERA (Sovereign Experience Replay & Amplification)
• Surfaces what you learned — nl digest shows KB patterns grouped by domain
• Compounds — each nightly run builds on the last, 116+ nights of distillation already running

60-Second Quickstart

pip install neuraloop

nl up                    # starts proxy on port 5055
nl setup-mcp --all       # writes MCP config for Cursor, Windsurf, Claude Code
nl digest                # see what SERA learned about your codebase
nl intel matrix          # competitor intelligence surface

Why Not the Alternatives

	Neuraloop	epic.dev	Mem0	OneContext
Self-improving	SERA	manual	no	no
Local-first	yes	cloud	cloud	cloud
MCP-native	37 tools	yes	no	no
Nightly distillation	yes	no	no	no
Air-gap capable	yes	no	no	no

The SERA Moat

SERA runs 12 phases every night:

observe → extract → curate → knowledge → dream → synthesize
       → distill → evolve → promote → report → compress → schedule

Each run produces gold patterns (quality >= 0.7) from your actual session history. Pattern 82 (run 116) is informed by patterns 1-81. The corpus compounds. A competitor starting today cannot catch up — the gap widens every night.

Architecture

IDE Stop Hook → Shadow Stream JSONL → ChromaDB (22k+ vectors)
                                           |
                              SERA Nightshift (03:00 AM)
                                           |
                        KB Gold Patterns (~82 patterns, 5 domains)
                                           |
                              nl digest → proactive surfacing

Proxy runs locally on port 5055. All data in ~/.gemini/antigravity/. Zero telemetry. Zero cloud calls. Ollama embeddings via local nomic-embed-text.

Commands

nl up              # start proxy + SERA engine
nl digest          # show KB patterns by domain (last 7 days)
nl intel matrix    # competitor threat matrix
nl intel gaps      # 3 structural gaps vs competitors
nl setup-mcp       # write IDE MCP configs (cursor/windsurf/claude-code)
nl status          # system health: proxy, ChromaDB, SERA, KB
nl pulse           # live event stream from shadow stream
nl architect       # generate architecture from SERA KB patterns

Requirements

• Python 3.11+
• Ollama (for embeddings): ollama.ai
• 4GB RAM minimum (ChromaDB + Ollama)

License

MIT