AI SupportConceptResolvo
An AI support agent that knows when to ask a human.
An AI customer support agent that answers from approved knowledge, drafts replies, takes actions such as refunds within limits and hands off to people when it should.
- Category
- AI · Customer Support
- Designed for
- Web
Why Resolvo exists
The problem
Support teams drown in repetitive tickets, and chatbots either say nothing useful or make things up and promise refunds they should not.
The product
Resolvo answers from approved knowledge with citations, acts within limits you set, and hands off to people with the context already gathered.
Built for: Support teams at SaaS and e-commerce companies handling 1,000 to 100,000 conversations a month.
- Answers grounded in approved knowledge, with citations
- Actions such as refunds with approval thresholds
- Handoff rules for sensitive or low-confidence cases
- Knowledge gap and conflict detection
- Red-team test suite for prompt injection
- Audited resolution rate and cost per resolution
Walkthrough
AI That Earns Autonomy Step By Step
Designed so support leads decide exactly what the agent may do alone, and can see why it did it.
- 01
Inbox With AI Drafts
Conversations needing a person, with the AI's draft, proposed action, sources and confidence ready to approve.
- 02
Knowledge Sources
Synced help centre, policies and docs with coverage, unanswered questions and conflicting answers flagged.
- 03
Actions & Guardrails
What the agent may do automatically, what needs approval, when it must hand off and what it must never do.
- 04
Performance
Audited AI resolution rate, satisfaction, first response time and cost per resolution by week.
Every screen
The full experience.
What makes it different
Grounded, Limited, Auditable
Answers come only from approved sources, actions go through scoped tools with approval thresholds, and every conversation is traceable.
Grounded Answers
Retrieval over the help centre and policies, with citations on every reply.
Scoped Actions
Each tool has its own permission, limit and approval rule.
Clean Handoffs
Clear rules for when a person takes over, with a summary so customers never repeat themselves.
Tested Before Release
Every change runs against policy and prompt-injection test suites.
How it’s built
The technology stack.
The architecture we would build this on today, and why each piece earns its place.
- Web app
Next.js, React, TypeScript, Tailwind CSS
Review queues and dashboards that stay fast with thousands of items
- AI services
Python (FastAPI) · Claude via the Anthropic API with tool use
Strong reasoning and structured tool calls, isolated from the core API
- Tools & integrations
Model Context Protocol (MCP) servers
One standard way for agents to reach business systems, with scoped permissions
- Durable workflows
Temporal
Long-running, multi-step AI work that survives restarts and waits for approvals
- Knowledge & memory
PostgreSQL + pgvector
Retrieval over approved sources without a separate vector database
- Help desk
Zendesk, Intercom or Freshdesk APIs
Works inside the tools the support team already uses
- Evaluation & tracing
Versioned eval suites, OpenTelemetry traces, cost tracking
Every prompt or model change is tested before release and every run is explainable
More inside
The Support Desk, Augmented.
The screens a support lead uses to run an AI agent safely.
Test Suites
Replay real conversations and adversarial cases before any change goes live.
Policy Controls
Topics and promises the agent will never make, enforced before a reply is sent.
What comes next
On the roadmap.
What a second release would add once the first one is in real users’ hands.
- 01Voice channel support
- 02Proactive outreach for known incidents
- 03Multilingual answers from one knowledge base
- 04Customer-facing help widget
- 05QA scoring of human replies
Related
Services and industries behind this work.
Concept project
Need Something Like This?
This is concept work: the product design, screens and architecture we would start from. If your business needs an AI support agent connected to your own systems, we can scope a first release with you, usually within a week.
Next project
Docuvia
AI that reads invoices, checks them and posts them.
