Skip to content
Grounded answers, no hallucinations

Turn your documentation into a support agent that actually knows your product.

Support Agent answers customer questions only from your approved docs and Q&A — and clearly says when it doesn't know. Bring your own model, connect your inbox, and keep a human in the loop with Slack approvals.

Get startedSign in
  • Bring your own key
  • Self-hostable
  • Your data, your Postgres
Support agent · live
How do I reset my API key?
Open Settings → Security, then choose Rotate API key. Your old key keeps working for 24 hours so you can migrate safely.
Sourcesapi-keys.mdsecurity-faq
Answered from approved content2 chunks retrieved

Works with your model — bring your own key

  • OpenAIGPT-4o, o-series
  • AnthropicClaude
  • GoogleGemini
  • Local / self-hostedOpenAI-compatible

Retrieval is app-managed, so even small models answer accurately — they only read the context we hand them.

The problem

Generic chatbots make things up.

Drop a general-purpose model in front of customers and it confidently invents features, pricing, and steps that don't exist. Every wrong answer is a support ticket, an angry customer, and lost trust.

Customer: “Does the Starter plan include SSO?”
Generic bot: “Yes! SSO is included on all plans.” (It isn't.)

The solution

Answers grounded in your approved content.

Support Agent retrieves the right passages from your docs and Q&A, answers only from them, and falls back gracefully when the knowledge base doesn't cover the question — no guessing.

  • Retrieves the most relevant chunks from your indexed content
  • Answers strictly from retrieved context and live tool results
  • Says "I couldn’t find that" instead of inventing an answer

Features

Everything you need to ship an accurate support agent

Built around one principle: answer from approved content, and be honest when the answer isn’t there.

Grounded answers (RAG)

Upload docs or crawl your help site, add Q&A snippets, and we chunk, embed, and retrieve the right context so every answer is backed by your own content.

Any model, your keys

Power each agent with OpenAI, Anthropic Claude, Google Gemini, or a local/self-hosted model. Bring your own API key — encrypted, never shared.

Live data tools

Register your own API endpoints so the agent can look up per-user data — subscription, billing, account status — at answer time, not just static docs.

Email support inbox

Connect your own mailbox over SMTP/IMAP (Gmail app password included). The agent receives support email and replies from your own address.

Slack approvals

Keep a human in the loop. Drafted replies post to Slack with Approve, Edit, or Reject — nothing sends until someone signs off.

Analytics that improve docs

See top unanswered questions, your most-used documents, and fallback rate so you know exactly where to strengthen your knowledge base.

How it works

From documentation to a live agent in four steps

1

Add knowledge

Upload .md, .txt, .docx, .pdf, or .html files, crawl your help site, or paste Q&A snippets. We chunk, embed, and index everything for retrieval.

2

Pick your model

Choose OpenAI, Anthropic, Gemini, or a local model and add your own key. Retrieval is app-managed, so small models work too.

3

Connect channels

Answer in-app, or wire up your mailbox so the agent replies to support email. Turn on Slack approvals to review drafts before they send.

4

Go live & learn

Publish your agent and watch the analytics — unanswered questions and fallback rate show you exactly which docs to improve next.

Accuracy by design

Answers only from approved content. It says when it doesn't know.

No hallucinated features, prices, or policies. Tools return live data; everything else is grounded in your knowledge base.

Only approved content

Answers come from your indexed docs and Q&A snippets — never from the open internet or the model’s imagination.

Live data, when it’s real

For per-user questions the agent calls your registered endpoints and treats the result as authoritative — it never fabricates account data.

Honest fallbacks

When nothing relevant is retrieved and no tool can help, it says it couldn’t find the answer — and can escalate for human review.

Own your stack

Self-host it. Your data stays yours.

Support Agent is built to run on your infrastructure with a single Docker stack — so customer questions, documents, and credentials never depend on a third party.

Your keys

Provider API keys and SMTP credentials are encrypted at rest and never leave your deployment.

Your Postgres

Documents, embeddings, and conversations live in your own PostgreSQL with pgvector — no external vector database.

Your infrastructure

Ship with Docker and run the whole stack on your own servers. Nothing phones home.

FAQ

Questions, answered

Everything you need to know before you wire up your first agent.

Which file types can I upload?
You can upload Markdown (.md), plain text (.txt), Word (.docx), PDF (.pdf), and HTML (.html) files. You can also crawl your existing help site, and add Q&A snippets directly in the app. Everything is chunked, embedded, and indexed for retrieval.
Which model providers are supported?
OpenAI, Anthropic Claude, Google Gemini, and any local or self-hosted OpenAI-compatible model. You choose the provider and model per agent and bring your own API key. Because retrieval is app-managed, even small models answer accurately.
Does it hallucinate?
It's designed not to. The agent answers only from the context retrieved from your approved docs and Q&A, plus the results of any tools it calls. When nothing relevant is found, it tells the customer it couldn't find the information rather than guessing.
Can it look up live, per-user data?
Yes. Register your own API endpoints as tools — for example a subscription or billing lookup — and the agent calls them at answer time with validated arguments. Tool results are treated as authoritative, and the agent never invents that data.
Can it reply to support emails?
Connect your own mailbox over SMTP/IMAP (including Gmail with an app password). The agent receives inbound email, threads it into a conversation, drafts a grounded reply, and sends it from your own address — optionally after Slack approval.
Do you store our API keys?
Your provider keys and mailbox credentials are encrypted at rest in your own database and are never exposed to the model, end users, or us. When you self-host, they never leave your infrastructure.
Can we self-host?
Yes. Support Agent ships as a Docker stack and runs entirely on your own servers, with your data in your PostgreSQL (with pgvector). There are no external vector databases and nothing phones home.

Give your customers answers they can trust.

Connect your docs, bring your model, and ship a support agent that knows your product — in an afternoon.

Get startedSign in