The problem
Support teams want AI to answer customer questions, but generic chatbots guess. They give answers nobody can verify, and they may take actions (refunds, account changes) with no oversight. Teams need AI that answers from their documentation, shows where each answer came from, and asks before doing anything risky.
The solution
ResolveAI is a multi-tenant support workspace. Teams upload their own knowledge, and the AI answers from it with citations. Every agent run is logged step by step, and risky "write" tools pause until a human approves them. The flow is: Ingest → Index → Resolve. Documents are chunked and embedded, then retrieved to ground each answer.
Key features
- Private knowledge base. Upload PDF, Markdown, DOCX and TXT. Documents are chunked, embedded, and found with hybrid search.
- Grounded chat with citations. Each answer links to its source document chunks with relevance scores.
- Multi-agent resolution pipeline. Triage → Retrieval → Diagnostic/Tools → Resolution/QA, saved as an inspectable agent-run timeline.
- Human approval gates. Risky tool calls (e.g. account writes) wait for Approve/Reject and are audit-logged.
- Workspace security. Tenant isolation, RBAC (Owner, Admin, Support, Developer, Viewer), JWT access with hashed refresh tokens, email verification, session revoke, and audit logs.
- Usage controls. Free plan with per-workspace limits (100 AI requests/day, 100K tokens/day, 1M tokens/month), integrations and usage analytics.
- Self-serve onboarding: register → verify email → upload → ask. Docs for the API and Postman collections.
Stack
Next.js front end (Turbopack build), with Sora/Manrope/JetBrains Mono typography and a light/dark theme toggle. The site mentions JWT auth and embeddings/vector retrieval.