Letting AI read and write enterprise systems means every operation can affect real data. Security isn't an add-on — it's the platform's foundational architecture.
Each user can only access their own conversations, files, and knowledge bases. Uploaded files are stored with user-level isolation. SSE streaming supports token authentication for secure real-time communication. First launch guides admin account creation, no config file editing needed.
Agent auto-pauses before executing data modifications, approval initiations, and similar operations, sending confirmation requests to designated personnel.
Complete record of every operation with export: timestamp, user, connector, Action, parameters, response. Admin review log audit tab for reviewing all platform activity. Supports conditional filtering and export, meeting classified protection and compliance audit requirements.
Full organization system with CRUD and role-based membership (owner, admin, member). Three-tier resource visibility: personal, organization-shared, and global. Publish and unpublish resources via API, with admin management UI for complete organizational control.
Built-in platform organization that auto-joins all users. Serves as the central platform for sharing resources across the entire organization: connectors, agents, knowledge bases, and more are accessible to everyone through this shared space.
Users browse and subscribe to shared resources published by organizations. Subscribe and unsubscribe via UI or API. All resource types (connectors, agents, knowledge bases) support org-level publishing for team-wide sharing.
Review toggle per organization. When enabled, published resources go through an approval workflow with a review sheet (approve/reject). Status badges on resource cards show review state at a glance.
Three registration modes: open (anyone can sign up), invite-only (requires an invite code), and disabled (no new registrations). Full invite code CRUD for managing access: create, distribute, and revoke invite codes as needed.
Benchmark agents before deployment. Run evaluation suites against agents to measure accuracy, latency, and reliability, ensuring production readiness before going live.
SQLite for zero-config default setup, PostgreSQL for production deployments. Docker Compose auto-provisions PostgreSQL. No manual database setup required. Migrate seamlessly from SQLite to PostgreSQL as your deployment scales.
Compatible with any provider supporting OpenAI /v1/chat/completions interface: OpenAI, Anthropic, DeepSeek, Qwen, Ollama, vLLM, etc. Switch by changing LLM_BASE_URL and LLM_MODEL, no business logic changes needed.
