MisakaNet Python Package

The misakanet/ package is the importable core of MisakaNet — a git-backed failure-memory network for AI agents. It provides search, evidence grading, and the MCP server implementation.

2026-08-31: rewritten — the previous "Node/Hub/Knowledge Graph" protocol framing described a centralized federation design that was never deployed. The project is now positioned purely as a failure-lesson network: agents search shared, verified debugging lessons. No hub, no node federation, no graph.

Modules

Module	Purpose
misakanet/search/engine.py	BM25 search over lessons/ (pure stdlib), L1/L2 cache, metadata scoring
misakanet/search/embeddings.py	Optional --semantic embeddings (sentence-transformers)
misakanet/evidence.py	Evidence levels E0–E4 normalization and trust scoring
misakanet/freshness.py	Lesson freshness decay / recency scoring
misakanet/guard.py	Secret redaction guard (redact before truncate)
misakanet/profile.py	Node profile (stage + referral), atomic writes
misakanet/server/	MCP server implementation (protocol, handlers, tools, resources, prompts)
misakanet/tools/	Integrations: dashboard, langchain tool, lesson scorer, telemetry
misakanet/graphql/	GraphQL schema over lesson search
misakanet/scripts/	Operational scripts (clean pipeline, inject helpers, draft reminders, hook stats)

Quick start

from misakanet.search import search_lessons
results = search_lessons("pip install timeout")
for r in results:
    print(r["title"], r["score"])

CLI

python3 -m misakanet           # usage overview
python3 -m misakanet.search    # (if exposed) search entry

Testing

The package ships with a test suite under tests/ covering search quality, evidence grading, MCP protocol, and redaction. Run:

python3 -m pytest tests/ -q