Memstate AI

What Memstate AI does

Memstate AI is a hosted agent-memory backend that organizes facts under hierarchical keypaths and retains a version chain. An agent can request the current value, inspect prior values, and surface a conflict when a decision changes. Its MCP package exposes remember, set, get, search, history, and soft-delete tools for Claude, Cursor, Cline, Windsurf, and related clients.

Official pricing lists Free at $0 for 100 memories and 2 projects; Pro at $29 monthly for 25,000 memories and 15 projects; Scale at $199 for 250,000 memories, unlimited projects, and up to three seats; Enterprise is custom with self-hosting. Define what one memory means because a summary may become several keypaths. The vendor publishes benchmarks against Mem0, but reported metrics differ between the README and current site. Treat them as hypotheses: rerun the open benchmark with pinned models and your own project histories, including reversals, ambiguous facts, deletion, and cross-project isolation.

Pricing, alternatives, and fit

Published pricing is: Free is $0; Pro is $29/month; Scale is $199/month; Enterprise is custom-priced with self-hosting. Pricing should be evaluated as total cost per accepted result, not only the headline fee. Include model calls, external APIs, compute, storage, retries, human review, implementation, and support. Run a capped pilot with representative peak volume and record successful completions, p50 and p95 latency, correction minutes, failures by cause, and spend per approved output.

Relevant comparisons are Cognee, Chroma, Claude, Cursor. Compare them on the same inputs because category labels hide differences in hosting, permissions, integrations, observability, and exportability. The best choice is the one that meets a written quality and recovery threshold on the buyer's own workload, not the one with the longest feature list.

Pros, cons, and implementation guidance

Concrete strengths include Changed decisions stay auditable; Keypath retrieval avoids irrelevant similarity matches; MCP works with several coding clients; Pricing states memory and project limits. Important limitations include Benchmarks are vendor-produced; Extraction can store an incorrect interpretation; Hosted memory centralizes sensitive context; Open MCP client does not make the backend open source. These tradeoffs need owners. Security review should cover SSO, role-based access, encryption, secrets, audit logs, subprocessors, retention, deletion, incident response, and whether customer data trains shared models. Technical review should test rate limits, timeouts, retries, idempotency, concurrency, versioning, partial failure, export, and recovery after interruption.

Practical use cases are Preserve architecture decisions; Maintain current project facts; Detect revised requirements; Share memory across MCP clients. Start with one bounded workflow whose current time, cost, and error rate are known. Use at least 50 real examples plus malformed and adversarial inputs. Keep a manual fallback until the system repeatedly meets the acceptance bar. For tools that execute actions, use least privilege and require human confirmation for money movement, publication, deletion, account changes, and external messages. For generated media or extracted documents, preserve provenance and source references so reviewers can verify consequential output.