Zero LLM Ingestion Tax · Local, Cloud, Private VPC & Air-Gap
Most memory tools were built as cloud wrappers for consumer chatbots, burning paid LLM tokens on every write while waiting passively for an agent to remember a tool call. Sulcus is a native Rust Semantic VMMU that curates, pages, and coordinates context automatically across your entire engineering fleet.
Mem0 wraps external LLMs and vector stores for app chatbots, while Sulcus runs a zero-LLM-tax Rust Semantic VMMU built for coding agents and engineering teams.
Every core capability side by side across Sulcus, Mem0, Zep, Supermemory, Honcho, and Letta.
| Capability | Sulcus | Mem0 | Zep | Supermemory | Honcho | Letta |
|---|---|---|---|---|---|---|
| Zero-LLM Ingestion Tax | Local CPU ONNX + SIU ($0) | Paid LLM calls per write | Paid LLM extraction per episode | Metered cloud processing credits | Background Deriver LLM burn | Agent self-edit + sleep LLM turns |
| Map vs Territory Paging | Delta pointers first, fetch on fault | Flat extracted fact strings | Context block assembly | Profile + RAG chunks | Peer card + dialectic query | Core vs archival blocks |
| Thermodynamic Decay | Continuous heat, boost, and decay | LLM overwrite / delete rules | Bi-temporal edge timestamps | Recency weighting | Append-only reasoning queue | Agent-managed block edits |
| Knowledge Graph Included | Every tier, including Free | Locked behind $249/mo Pro | Core Graphiti architecture | Document chunk links | Conversational peer representations | Hierarchical MemFS files |
| Proactive IDE & CLI Hooks | 15+ harnesses (VS Code, Junie, CLI) | Passive OpenMemory MCP | Passive Context MCP | Passive OAuth MCP | SDK / API integration only | Letta Code harness only |
| Keep Existing Coding Agents | Works with every IDE and CLI | External SDK / MCP | External SDK / MCP | External SDK / MCP | External SDK / API | Replaces runtime with Letta |
| Live Fleet Comms & File Safety | SSE Bus + territory collision alerts | None | None | None | Multi-peer chat state only | Inside Letta server only |
| Surgical Code + Memory Undo | Pre/post file + graph snapshot rewind | None | Temporal edge history only | None | None | Git MemFS for memory only |
| Two-Gate Namespace RBAC | Explicit Read / Write / Deny matrix | Enterprise custom tier | Project / user isolation | Org container tags | App / peer scoping only | Cloud org workspaces |
| 10-Seat Private VPC / Air-Gap | $2,400/yr + offline Control Panel | Custom Enterprise quote | $4,500+/yr or DIY Neo4j + LLM | $4,788/yr Scale or custom quote | Custom deal or DIY Python + LLM | Custom Enterprise quote |
Competitors fire GPT or Claude calls on every turn just to extract facts or graph triples. Sulcus runs local CPU ONNX embeddings and SIU curation in under 50ms, so memory writes don't inflate your model bill.
Passive MCP servers fail whenever an agent forgets to call search or store. Sulcus hooks into IDE and CLI lifecycles to inject delta Map pointers before turns start and distill verified outcomes when turns complete.
Run one local Rust binary free forever, sync across Sulcus Cloud, or deploy inside your own AWS VPC or air-gapped enclave with an embedded 8-tab Enterprise Control Panel and cryptographic seat governance.
Start free on your workstation in minutes, or roll out governed shared memory across your engineering organization.