Comparisons
Statewave vs. Mem0, Letta, Zep & Supermemory
How Statewave's deterministic, token-bounded context assembly compares to the other memory systems teams evaluate alongside it.
Same question, four times: does the system compile a fixed answer ahead of time, or search and re-rank one at call time? The matrix below is the short version; each card links to the full breakdown.
| Product | How context is selected | Deployment |
|---|---|---|
| Mem0 | Top-k embedding-nearest as the candidate pool, then re-scored | Apache 2.0 core, paid platform |
| Letta | The agent reads and edits its own context tree with tool calls | Apache 2.0, with managed Letta Cloud |
| Zep | Graph traversal plus a managed reranker | Cloud, BYOK, or BYOC — self-hosted Community Edition discontinued 2025 |
| Supermemory | Hybrid vector + keyword search with a context-aware reranker | Self-hostable binary (10k-doc cap stated for the local build since v0.0.7) or a managed hosted platform |
| Statewave | Deterministic assembly, ranked to a token budget | Apache 2.0 throughout, runs fully offline |
Statewave vs Mem0
Deterministic Context vs. Ranked Retrieval
Mem0 ranks memories by relevance for an id you pass. Statewave compiles episodes into typed memories with a fixed scoring model, policy on the read path, and an optional integrity-hashed receipt.
Read the comparisonStatewave vs Letta
Runtime-Managed vs. Agent-Managed Memory
In Letta the agent edits its own memory blocks with tool calls. Statewave assembles a deterministic, token-bounded bundle mechanically, with no model in the loop.
Read the comparisonStatewave vs Zep
Inspectable Bundle vs. Opaque Context Block
Zep models memory as a knowledge graph and returns an opaque Context Block string. Statewave returns typed, provenance-traced memories with per-row confidence and validity.
Read the comparisonStatewave vs Supermemory
Deterministic Bundle vs. Reranked Search
Supermemory does hybrid vector-plus-keyword search with context-aware reranking. Statewave assembles a deterministic bundle with per-row provenance and an optional integrity-hashed receipt.
Read the comparisonOpen-source alternatives, by product
Longer comparisons that go beyond Statewave — six alternatives to each product, evaluated on their own merits.
6 Best Open-Source Supermemory Alternatives
Compare six Supermemory alternatives for governed agent memory, local context, temporal graphs, reflection, and self-hosting.
6 Best Open-Source Alternatives to Mem0
Mem0's self-hosted build lacks graph memory, webhooks, and audit logs. Compare six open-source alternatives on benchmarks, governance, and what's actually free.
FAQ
Frequently asked questions
How does Statewave compare to Mem0, Letta, Zep, and Supermemory?
All four rank or retrieve memory at query time in their own way — Mem0 by relevance score, Letta through agent-managed tool calls, Zep via knowledge-graph traversal, Supermemory with hybrid search plus reranking. Statewave instead compiles episodes into typed memories ahead of time and assembles a deterministic, token-bounded bundle with policy enforcement and an optional integrity-hashed receipt of what was delivered.
Is Statewave self-hosted like these alternatives?
Yes — Apache-2.0, self-hosted on PostgreSQL and pgvector, with no cloud dependency. That is not universal among the alternatives: Zep discontinued its self-hosted Community Edition in 2025 and now runs memory only through Zep Cloud, BYOK, or Bring-Your-Own-Cloud.
Are Mem0, Letta, Zep, and Supermemory open source?
Partially, and the gate lands in different places. Mem0’s core is Apache 2.0 but its hosted platform is paid. Letta is Apache 2.0 with a managed Letta Cloud. Supermemory’s repo is MIT, but its v0.0.7 release notes first stated a 10,000-document cap for the local build and supported self-hosted deployment starts on its $399/month Scale plan. Zep dropped its self-hosted Community Edition in 2025. Statewave is Apache-2.0 across the server and SDKs with no paid tier gating self-hosting.
Which comparison should I read first?
Match it to what you already have: read vs Mem0 if you’re replacing an existing Mem0 integration, vs Letta if your agent framework already manages its own memory with tool calls, vs Zep if you need a knowledge-graph model or are affected by its Community Edition sunset, and vs Supermemory if you need the multimodal or local-binary surface it covers.
Answers last checked against the Statewave docs and repositories on .
Open benchmark
Comparing something else?
Statewave is open source and self-hosted, so you can reproduce our published comparison yourself, or use the same harness as the starting point for benchmarking your own memory system.