Standard AI dashboards give you a delayed, total bill. They don't tell you which agent, what model, or why a runaway loop just cost you $50 in 10 minutes.
See every API request as it happens — provider, model, token count, latency, and cost. No more waiting for end-of-month surprises.
Assign isolated virtual keys to specific agents, apps, or team members. Track exactly who spent what and on which provider.
Set hard budget limits per virtual key. When spend hits the threshold, the key stops working — preventing runaway agent costs.
Choose the path that fits your security requirements. Same features, different hosting.
Everything you need to manage multiple AI providers, agents, and projects from one dashboard.
Works with OpenAI, Anthropic, Google Gemini, Mistral, DeepSeek, and xAI (Grok) — 6 providers out of the box. Just change your BASE_URL — the proxy handles auth, logging, rate limiting, and cost tracking automatically.
Register AI agents (OpenClaw, custom scripts, any autonomous system) and track them with heartbeats. Real-time status: healthy, degraded, or offline.
Create isolated keys with budget limits, rate limiting, model allowlists, and expiration dates. Rotate keys without touching your app code.
Group agents, keys, and usage by product. See per-project cost breakdowns and resource allocation at a glance.
Get alerts when keys hit budget limits or agents go offline. Deliver them via email (Resend), Slack webhooks, or custom webhooks with HMAC signatures.
Multi-org support with role-based access control: owner, admin, and member roles. Invite team members with invite codes.
AI-powered model recommendations that analyze your usage and suggest cheaper alternatives — same tier or budget options. See exactly how much you can save each month.
Track money lost to failed requests. Error heatmaps, per-key spend drilldowns, and budget suggestions help you eliminate waste and optimize your AI spend.
A modern stack designed for speed, security, and developer experience.
Framework
Next.js 16
React 19, App Router
Database
PostgreSQL
Drizzle ORM, Supabase
Encryption
AES-256-GCM
PBKDF2, 100k iterations
Auth
Auth.js v5
JWT, OAuth, Credentials
Proxy Overhead
< 5ms
Minimal latency added
Docker Image
Alpine-based
Node.js 20, lightweight
AI Providers
6 Built-in
OpenAI, Anthropic, Google, Mistral, DeepSeek, xAI
Testing
147 Tests
Vitest, full coverage
Every API request passes through a 7-step security and logging pipeline.
Virtual key extracted from Bearer token and verified against SHA-256 hash.
Check key is active, not expired, within budget, under rate limit, and model is allowed.
Linked provider credential decrypted (AES-256-GCM in self-hosted, Supabase Vault in cloud).
Request forwarded to the provider (OpenAI, Anthropic, Google, Mistral, DeepSeek, or xAI) with real API key injected.
Tokens, latency, model, cost, and status logged asynchronously — zero impact on response time.
Budget thresholds checked. If spend hits 80% or 100%, alert created and delivered to your notification channels (email, Slack, webhook).
Full provider response returned with X-RateLimit headers. No data modified, no lock-in.