Your AI tools forget everything. Rebis remembers.
A self-optimizing intelligence engine that runs on your hardware, compounds across every project, and cuts wasted AI token costs by eliminating redundant context. Works across every LLM. Your data never leaves your machine.
Not another memory tool
Commodity memory systems store and retrieve. Rebis is structured intelligence that activates, challenges, and compounds. Use focused context to reduce repeated setup and retrieval.
Structured intelligence
Not a list or vector database. Connected neurons encode how knowledge relates—when one insight is relevant, the full context fires together. Structure becomes intelligence. Your AI gets complete understanding, not isolated search results.
Adapts to you
Rebis is a personalized learning model that adjusts based on real-world experience—your experience, not a one-size-fits-all dataset. It captures what works and what doesn't, weighting both equally. Every user's brain is unique because every user's workflow is unique.
Blind spot detection
Independent oversight watches for over-reliance on familiar patterns, stale knowledge, and coverage gaps. Your brain actively challenges its own recommendations before they reach you—structural defense against tunnel vision and confirmation bias.
Compounding returns
Each session adds depth. Knowledge that proves useful gains weight. Knowledge that goes stale fades naturally. The result is a brain that gets measurably sharper with every session—compounding intelligence, not accumulating data.
Watch your brain think
This is a real subgraph from a working Rebis brain. Hover neurons to explore. Click "Fire" to watch intelligence cascade through connected knowledge.
Multiple systems, not a single trick
Rebis doesn't just retrieve. Distinct layers work together so your AI tools give better answers, avoid blind spots, and get smarter with use.
Instant Recall
Your current task instantly activates the most relevant knowledge. Related neurons fire together—you get a complete picture, not isolated facts. The entire brain responds to your context in milliseconds.
Result: API Security → Privacy Controls → Memory Safety → Data Isolation
These fire as a coherent cluster, not individual lookups.
Self-Challenge
An independent layer watches for what the recall system might miss: coverage gaps, over-reliance on familiar patterns, and stale knowledge. The brain checks its own recommendations before they reach you—catching blind spots that no retrieval system would flag.
Cross-Pollination
Deliberately surfaces knowledge from unrelated domains that might apply by analogy. Generates fresh questions that challenge conventional thinking—connections you wouldn't have made yourself, drawn from the full breadth of your accumulated experience.
Signal Focus
When many neurons activate with similar relevance, competitive ranking ensures only the highest-quality recommendations reach you. No information overload—just focused, actionable intelligence. The rest is available on demand.
Continuous Growth
Record what worked, what failed, and what was surprising through the trace tools. Maintenance can use these records to update the brain; capture depends on your integration and workflow.
Three steps to a brain that compounds
Install a release, configure your client, and record useful knowledge as you work.
Install
Install the Python package, initialize a brain, and configure your tool connection. MCP and HTTP integrations require client configuration; release availability varies by platform.
install → init → connect to your tools
# Full setup guide included after signup
Your brain grows
Capture reasoning traces through a configured integration. Review and maintain your knowledge as you move between projects and tools.
Session 1: +3 neurons (auth patterns)
Session 5: +8 neurons, connections form
Session 20: 47 neurons, graduated trust
Session 100: domain expert
AI gets smarter
Context compounds. Your AI stops asking the same questions and starts building on what it already knows.
Session 2: "What framework?"
Session 3: "What framework?"
Session 2: Applies + learns deploy
Session 3: Suggests optimizations
Research into brain-powered orchestration
The repository includes multi-agent federation experiments and a simulation study of domain-aware coordination. These research results describe simulated behavior; they do not establish a complete production fleet service.
Exploring domain-aware leadership
The research models agents with separate knowledge graphs and compares task-specific leader selection strategies. Domain depth and model capability are inputs to the simulated election. Connecting these ideas to a deployed agent fleet requires additional integration and validation.
Illustrative research scenario: Your Claude agent has been doing backend work for months while a GPT agent handles data analysis. A new task arrives: “Optimize the analytics queries.” A domain-aware election could score both brains and select a lead. This example describes the study design, rather than a turnkey deployment.
- Domain-depth election: composite scoring achieves 92.4% leader accuracy (ω²=0.352, large effect)
- Key finding: partial coordination is worse than none—commit fully or skip it (ω²=0.072)
- Scale-dependent configuration: election×fleet interaction ω²=0.163 (largest effect in study)
- Quality and resilience emerge together: r=0.485 correlation between task quality and failure recovery
- Multi-model routing: election factors in domain knowledge and LLM capability match—not just model size
Research scope: Federation code and experiments are included. Production identity, transport security, and fleet operation require deployment-specific integration.
Built on data science, not defaults
6 independent variables × 7 outcome measures. Full factorial design across 486 unique configurations, 500 replications each. Non-parametric statistical methods with Benjamini-Hochberg FDR correction at q=0.001, effect sizes with confidence intervals, and achieved power of 1.0 across all tests. Simulation findings inform research; production performance and configuration require separate validation.
Read the 243K-trial study abstract →
MCP Bridge — connect compatible clients
The MCP Bridge exposes a configured Rebis brain for compatible clients. Client support and transport setup vary. Retrieved context sent to a hosted model is subject to that provider's data handling; storing the graph locally does not keep all model inputs local.
One brain. Every LLM. At the same time.
Rebis doesn't belong to any LLM. It works across all of them—simultaneously.
Priced for real value, not volume
A mature Rebis brain saves multiples of its cost in reduced token spend and eliminated rework. Two dimensions: human seats and agent seats. Self-hosted. No cloud bill.
- Full LOBE runtime
- Core activation and trace tools
- MCP + HTTP integration
- Priority updates + support
- Neuron Pack marketplace
- Advanced trace analytics
14 days free. Then $69/mo. Cancel during trial—no charge. Cancel after—access runs through your paid period, no partial refunds.
- Everything in Developer
- Federation experiments for team workflows
- Signed envelope routing (mTLS)
- Priority support + SLA
- Everything in Team
- Dedicated architecture reviews
- Custom Neuron Packs
- SSO + compliance tooling
Agent & Bot Seats
Autonomous agents use Rebis differently—higher volume, always-on, more activation calls. Per-agent pricing or fleet packs for multi-bot orchestration.
Each agent seat can use activation and trace tools. Integration with an orchestration framework requires configuration and testing.
Start Subscription| Pack | Price | Per Agent | Savings | |
|---|---|---|---|---|
| 5 agents | $159 / mo | $31.80/agent | 18% off | Buy → |
| 25 agents | $599 / mo | $23.96/agent | 39% off | Buy → |
| 100 agents | $1,900 / mo | $19.00/agent | 51% off | Buy → |
| 250+ agents | Custom | Volume pricing | Contact us |
Connect any LLM via MCP Bridge
The MCP Bridge offers an additional connection path for compatible clients. Check the client's supported transport and configure access before connecting. Context passed to hosted services leaves your machine.
Install → Initialize → Bridge starts
# Your brain is now available to
any MCP-compatible client over SSE
- MCP over SSE (standard protocol)
- Optional bearer token auth
- Localhost-only by default
- Full setup guide included with license
- Claude Code / Desktop — native MCP (stdio)
- Claude.ai — via MCP Bridge
- ChatGPT — via MCP Bridge
- Perplexity — via MCP Bridge
- Custom apps — HTTP API or Bridge
- SaaS copilots — embed via Bridge
- Bot fleets — per-agent Bridge instances
The moat: Even if LLMs build their own memory, that data lives on their servers. With the MCP Bridge, your knowledge stays on your machine. Ownership, not rental.
Managed cloud hosting is on the roadmap. Interested? Let us know.
Zero vendor custody by design
Rebis stores the knowledge graph in your chosen environment. Configure storage protection and backups for that environment. Only connect tools you trust; context shared with hosted models follows their data policies.
Start building a brain that compounds
Your AI tools will never start from zero again.
Not ready yet? Get product updates and research findings.
No spam. Unsubscribe anytime.
You're in. We'll be in touch.