Case studies
03 projectsResult
Fully self-hosted via Docker Compose or GKE. Replaces mem0, Zep, Nango, and Dify at ~$50–100/month for 10K items versus ~$260–280 on cloud APIs alone.
View on GitHub →Overview
A private AI memory platform that ingests data from 300+ apps via Nango-powered OAuth webhooks, enriches every item through a 42-agent AI pipeline, and stores it with versioning and per-item ACLs. Queryable through vector, BM25 full-text, hybrid, graph, and full multi-signal search — backed by a temporal knowledge graph built on Graphiti and Neo4j. Supports 10 memory types, automatic token refresh, MCP server integration for Claude and Cursor, and SDKs in Python, TypeScript, Go, Rust, and Ruby.
Discuss a similar project →Result
Deploys to GKE, EKS, or AKS via Helm and Kustomize. Per-user token tracking, cost calculation, and daily/monthly budget enforcement with a React 19 dashboard. Dynamic sync from HuggingFace and OpenRouter keeps the catalog current.
View on GitHub →Overview
A self-hosted model garden that manages 105+ open and public AI models with intelligent routing, backend scheduling, and multi-model task planning. Automatically routes requests to the best available model — preferring loaded models, then warm cache, then open source, then domain fit. Manages vLLM (GPU) and LocalAI (CPU) backends with a provider bridge unifying Anthropic, OpenAI, Google, Mistral, Groq, Together, Cohere, and Azure OpenAI behind a single interface.
Discuss a similar project →Result
Launched with three pricing tiers (Free, Plus $19/mo, Pro $39/mo). One user caught a water issue early using AI-detected condition score drops, preventing $5,000+ in damage. Insurance claims approved in 3 days using AI-verified PDF documentation.
Visit asset-mem.com →Overview
A complete property care platform where AI agents analyze timeline photos and documents, rate condition over time, flag issues, generate formal PDF reports, and guide users on repairs and costs while connecting them with local pros. Features six specialized agents (Timeline, Coverage, DIY, Service, Cost, and Report), two chat modes, before/after checkpoint comparisons, and multi-property management from a single account.
Discuss a similar project →