IntelligenceDesigned To Evolve
Build applications that reason, adapt and collaborate using a modular AI platform designed for production.
Build applications that reason, adapt and collaborate using a modular AI platform designed for production.
Groq for fast agent calls, Gemini for long documents. Models are auto-detected and failover is automatic.
Vector search plus a temporal knowledge graph, so answers cite sources and follow how facts changed.
Compliance, security, FinOps and architecture agents plan, use tools and check each other.
CNN feature maps for diagrams, layouts and telemetry heatmaps. Try it in the console.
Post-quantum key exchange, mTLS, encrypted vault, sandboxed scripts and audit logs.
Connect Claude Desktop or Cursor to query graph state and run pipelines.
Nginx serves the site and applies security headers.
FastAPI filters injection, redacts personal data and rate-limits by IP.
TF-IDF today; Qdrant and Neo4j in the full build.
Groq or Gemini answers. If one fails, the other takes over.
Groq, Gemini, LangGraph, PyTorch, Qdrant, Neo4j.
FastAPI, Pydantic, httpx, Fernet vault, hash-chained audit.
Docker Compose, Kubernetes with autoscaling, GitHub Actions CI and Pages.
Zero-knowledge proofs, Wasm sandbox, semantic cache, JWT roles.
Browser keys are held per tab by default and sent only to Groq or Gemini. With the backend, keys live in server environment variables.
The site can only call the two AI providers and your own API. No third-party trackers or analytics.
Prompt-injection filter, email and phone redaction, input caps and per-IP rate limits.
Every swarm and MCP action is written to a SHA-256 hash chain you can verify.
Non-root, read-only filesystem, no privilege escalation, Kubernetes seccomp profile.
Zero-knowledge proofs, Wasm sandboxing and JWT roles are on the roadmap, not shipped. See SECURITY.md.
Straight from your browser to Groq or Gemini, or to your own backend if you set a Backend URL. Never to us.
The code is MIT licensed. Groq and Gemini offer free tiers, so you only need free keys. Quotas apply.
It reads the live model lists from Groq and Gemini and picks a current general model, so it keeps working when providers retire models.
None by this site: no cookies, no analytics. Your messages go to the AI provider you configured.
Yes. Clone the repo and run docker compose up --build. Kubernetes manifests are included.
It has real safeguards, but the roadmap items above are not built. Review it against your own requirements before using confidential data.
Keys stay in your browser and go only to api.groq.com and generativelanguage.googleapis.com.
Four specialist agents (compliance, security, FinOps, architecture) plan together, then a judge scores the plan. Needs the Backend URL under Keys.
Paste a policy, config, log or code snippet. The AI returns a structured risk review with fixes. Uses your API (Worker or backend).
Runs a 3x3 Sobel convolution (the first layer of a CNN) fully in your browser. Nothing is uploaded.
Built in: strict Content-Security-Policy (only Groq and Gemini can be called), no-referrer, prompt-injection filter, email/phone redaction before sending, 600-character input cap, 20 requests per minute limit, text rendered as plain text (no HTML injection), keys never in the code.
Limits: a static site cannot hide a key from the person using it. Free-tier quotas apply, answers can be wrong, and the knowledge base is a small built-in sample. For shared or production use, move calls behind a server proxy (see the guide).