memledger 0.5.0a0

Memory governance and trust layer for AI agents — provenance, attribution, confidence gating

memledger

Memory governance and trust layer for multi-agent AI systems.

memledger adds attribution, provenance, confidence gating, and quality measurement on top of any vector-backed memory store. Every memory carries a source, a confidence score, a derivation chain, and an audit trail — so when agents share memory, trust transfers become legible instead of invisible.

⚠️ v0.5.0a0 — alpha release (name-reservation). Phase A + B of the v1 OSS launch are done; Phase C/D/E are in flight. pip install memledger (no --pre) will not pick this up — you must explicitly opt in (pip install memledger==0.5.0a0 or --pre). The stable v1.0.0 will follow when the launch checklist signs off. See CHANGELOG.md and docs/test-plan.md for what's verified vs. what's still on the runway. Do not use in production.

Why memledger

Existing memory frameworks (Mem0, Zep, Letta, AgentCore Memory) optimize for recall quality — how accurately the system retrieves relevant memory. None of them address accountability — who wrote this memory, how confident were they, what did they derive it from, and how has it been used since.

In a single-agent system, the accountability gap is academic. In a multi-agent system where agents read and act on each other's beliefs, it is the fault line where systems fail at scale.

What it does

• Attribution — every memory carries created_by, confidence, session_id, derived_from, supersedes, workflow_id, triggered_by, hedged, namespace
• Cross-agent lineage chains — derivation chains tracked across agent boundaries; effective confidence is the weakest link in the chain
• Confidence-gated retrieval — per-namespace PASS / FLAG / FILTER policy applied at search time
• Outcome feedback loop — record_outcome() updates memory confidence based on observed downstream outcomes
• Namespace RBAC — declarative per-agent access control over hierarchical namespaces
• Memory Attribution Integrity (MAI) evaluation — calibrated scorer (deterministic + LLM-as-judge) for memory quality
• OpenTelemetry observability — every operation emits spans with memory trust attributes
• MCP server — framework-agnostic adoption via Model Context Protocol

Install

pip install memledger

Backend extras (pick one or more):

pip install memledger[local,pgvector]   # zero-cloud OSS quickstart: local fastembed + Postgres+pgvector
pip install memledger[pgvector]         # bring your own LLM/embedding provider via LiteLLM
pip install memledger[sqlite]           # local SQLite (tests/demos only)
pip install memledger[aws]              # AWS path: Bedrock, OpenSearch SigV4, DynamoDB

Backend contract (v1): the only required backend is open-source PostgreSQL ≥ 14 with the pgvector extension ≥ 0.5. Aurora, RDS, Supabase, Neon and any other Postgres flavor work with the same backend class — only the DSN changes. See docs/architecture.md for verified deployment patterns (Docker, Kubernetes, AWS).

Quick start

from memledger import Memledger

ml = await Memledger.create(backend_name="sqlite", connection_string=":memory:")

await ml.add(
    content="HikariCP maxPoolSize=50 fixes payment-service OOM",
    namespace="/ops/incidents/payment-svc",
    confidence=0.9,
    created_by="ops-agent",
)

results = await ml.search(query="connection pool fix", namespace="/ops/incidents/payment-svc")
for r in results:
    print(r.content, r.confidence, r.created_by)

Architecture

System architecture, data model, and observability pipeline are documented in docs/architecture.md.

Companion repositories

memledger is split across three repositories under the memledger-ai organization:

• memledger-core — Python SDK, MCP server, Helm chart (this repo)
• memledger-ui — Trust graph visualization (FastAPI + React)
• memledger-docs — Documentation site (Docusaurus)

License

Apache 2.0. See LICENSE.