ContextGraph
ContextGraph is a governed shared-memory layer for multi-agent systems. It stores agent claims with provenance/freshness/trust metadata, enforces access control at retrieval time, and compiles token-budgeted “context packs” (optionally explainable) for each agent’s permissions. It provides a Python SDK, CLI/dashboard, an HTTP REST API, and an MCP server/tool integration, with optional Anthropic Claude Memory Tool adapter support and a Neo4j-backed self-hosted beta path.
Score Breakdown
⚙ Agent Friendliness
🔒 Security
Strengths indicated: policy-controlled retrieval (ACL/visibility), explainable filtering traces, provenance/freshness/trust/quality gates, and paid-claim locking to avoid content leakage across org boundaries. Unclear/undocumented: concrete authentication mechanism, TLS requirements for REST/MCP, key management/secret handling practices, scope model granularity at the API level, rate limiting, and detailed security posture (auditing details are referenced but not shown).
⚡ Reliability
Best When
You run multiple agents (and/or organizations) that must share knowledge safely, with auditable provenance, freshness/trust controls, and token-budgeted context assembly per caller permissions.
Avoid When
You only need lightweight retrieval and cannot justify governance overhead (provenance, reviews, sentinel verdicts, explainability, persistence of context packs).
Use Cases
- • Multi-agent teams needing reusable operational/research context without prompt glue
- • Governed retrieval with provenance, freshness/trust signals, and explainable inclusion/exclusion
- • Cross-agent/agent-to-agent workflows (support, incident ops, research handoffs) that require policy enforcement
- • Building governed RAG where memory is filtered by ACL/payment/quality gates
- • Operator workflows for sentinel-based claim validation and auditable verdicts
- • Claude/Anthropic-compatible memory operations backed by ContextGraph revisions and archival semantics
Not For
- • Personal single-chatbot memory with no governance/policy needs
- • Teams requiring a hosted enterprise IAM-first deployment (README emphasizes self-hosted/beta components)
- • Simple vector-database-only RAG pipelines without provenance or policy-controlled retrieval
Interface
Authentication
README describes ACL enforcement and agent permissions (e.g., org/partner/published visibility, freshness/trust gates) and governance endpoints, but does not document the concrete auth mechanism (API keys/JWT/OAuth) or whether scopes/roles map to API authentication.
Pricing
No pricing/subscription/tier information was present in the provided README/manifest content. Mentions 'optional payments' and 'paid claims' gating, but not pricing for the service itself.
Agent Metadata
Known Gotchas
- ⚠ Governed retrieval behavior depends on permissions (agent/org visibility) and claim gates (freshness/trust/sentinels/payment locks); agents may see empty/locked claims if authorization or gates don’t allow access.
- ⚠ Compiled context packs are persisted and retrievable; agents should use pack_id correctly rather than assuming recall/compile are stateless.
- ⚠ No explicit retry/idempotency semantics were documented in provided README content; agents may need to implement conservative retries for POST operations.
Alternatives
Full Evaluation Report
Comprehensive deep-dive: security analysis, reliability audit, agent experience review, cost modeling, competitive positioning, and improvement roadmap for ContextGraph.
AI-powered analysis · PDF + markdown · Delivered within 30 minutes
Package Brief
Quick verdict, integration guide, cost projections, gotchas with workarounds, and alternatives comparison.
Delivered within 10 minutes
Score Monitoring
Get alerted when this package's AF, security, or reliability scores change significantly. Stay ahead of regressions.
Continuous monitoring
Scores are editorial opinions as of 2026-03-30.