SuperLocalMemory
Information-Geometric Memory for AI Agents
SuperLocalMemory v3.6 "Optimize" is a local-first agent memory engine that also cuts your LLM bill: skip repeat calls at $0, compress prompts 60–95%, and stack with Anthropic’s 90% or OpenAI’s 50% provider KV-cache discount — all without a cloud proxy. Backed by three arXiv papers, EU AI Act compliant by architecture.
The Problem
Every AI agent memory system in production relies on cosine similarity over embeddings. At scale, this fails: flat similarity loses discriminative power, pairwise contradiction checking grows O(n²), and hardcoded lifecycle thresholds break on non-average workloads. Cloud-dependent architectures compound this with data residency risk and an August 2026 EU AI Act compliance deadline that engineering alone cannot resolve.
Key Capabilities
Fisher-Rao Geodesic Retrieval
Models each memory as a Gaussian distribution rather than a point. Distance is computed on the statistical manifold using the natural Fisher-Rao metric — confidence-weighted, improving with use.
Sheaf Cohomology Consistency
Detects contradictions in the knowledge graph algebraically via H¹(G, F). Catches global inconsistencies that pairwise checking cannot find, without O(n²) cost.
Riemannian Langevin Lifecycle
Replaces hardcoded thresholds with stochastic gradient flow on the Poincaré ball. Memory states self-organize to a provably optimal distribution based on actual usage patterns.
Three Operating Modes
Mode A: zero cloud, EU AI Act compliant, 74.8% LoCoMo. Mode B: Mode A + local Ollama LLM. Mode C: cloud LLM synthesis, 87.7% LoCoMo. Switch anytime.
17+ Tool Integrations via MCP
Works with Claude Code, Cursor, VS Code Copilot, Windsurf, ChatGPT Desktop, Gemini CLI, and over a dozen other AI tools out of the box via the Model Context Protocol.
v3.6 Optimize: Skip, Shrink, Discount
Local LLM cost-reduction layer. Exact-match cache skips repeat calls at $0. Structural compression shrinks prompts 60–95% before forwarding. Prefix alignment stacks with Anthropic’s 90% and OpenAI’s 50% native KV-cache discount. One install, no cloud proxy, no changed API keys.
Recall-Health Monitor (v3.6.8)
3-tier self-healing daemon: re-warm on cold start, readiness probe after each recall, circuit-breaker that resets the embedder when semantic channels degrade. Fixes the chronic silent BM25 fallback that caused undetected recall quality loss.
SkillFortify
Supply Chain Security for AI Agent Skills
Static analysis, behavioral sandboxing, and cryptographic attestation for the AI agent skill ecosystem. Detects malicious skills before they execute.
AgentAssert
Design-by-Contract for AI Agents
Formal specification and runtime enforcement of behavioral contracts for autonomous AI agents. Prevents drift, ensures compliance, enables composition.
AgentAssay
Regression Testing for Non-Deterministic AI Agents
Token-efficient stochastic behavioral testing framework purpose-built for non-deterministic AI agent workflows. Part of the Qualixar AI Reliability Engineering platform.