Production-grade RAG, built into your CRM
Turn your files, websites, docs and videos into a retrieval system built the way serious AI teams build them: hybrid search, re-ranking, cited answers, measured quality and an API. Your chatbot, voice agents and copilot all answer from it, and so can your own apps.
Knowledge base: Imago docs
Ask it anything about Imago, then open Answer quality and the API
This knowledge base is built from Imago's own docs. Ask about Imago, or tap a question. Answers come only from the sources, with citations.
Why this answer
Each answer shows the passages it used and how they scored on meaning, keywords and coverage.
Knowledge is included.
So is every other module, in one app.
- $0 to self-host, forever
- Your own keys, no markup
- Sandbox mode: try it with zero keys
Built like a production system
The numbers behind every answer.
40 + 40
Candidates per question
The top 40 by meaning and the top 40 by keywords, then re-ranked
HNSW
Vector indexes
pgvector inside PostgreSQL, next to your CRM data
p50 / p95
Latency and cost, measured
Scored through the real answer path against a random baseline
4 × 15
Settings Auto-tune sweeps
4 search methods, 1–15 passages, pool size, blend, chunk size, overlap
6
Safety tests before launch
Prompt injection, hidden instructions, prompt leaks, cross-business data, off-topic
0
Guesses
Below the minimum score it says “I couldn't find that” and logs the question
What you can do
Hybrid search that holds up
Meaning-based search (pgvector with HNSW indexes) and BM25 keyword search, fused and re-ranked, with neighbouring passages pulled in for context.
Smarter questions
Questions are rewritten and split into parts, so a messy customer question still finds the right passage.
Answers you can check
Numbered citations, quotes checked word for word against the source, and a clear “I couldn't find that” instead of a guess.
Measure before you ship
Test questions scored through the real answer path against a random baseline, with p50/p95 latency and cost per answer.
Auto-tune, with you in charge
Sweeps passages per answer, search method, pool size, blend, chunk size and overlap, then proposes changes you approve.
Safety tested
Prompt-injection and off-topic tests run before you go live, plus a production readiness check.
Any source, kept fresh
PDF, Word, CSV, Markdown, single pages, whole-site crawls, Docs pages and video transcripts, re-crawled on a schedule.
An API for your own apps
A REST endpoint with scoped API keys returns answers and citations, so any app or website can use the same knowledge.
Isolated and private
Each workspace's data is separated in the database itself, sources can be limited by team or role, and HIPAA mode is supported.
Questions about Knowledge
Which AI models does Imago support?
Anthropic, OpenAI, OpenRouter (one key for Claude, GPT, Gemini, Kimi, Qwen, Grok and more) and Ollama models on your own hardware. Every feature also runs in a sandbox with no keys.
How does the Imago knowledge base find answers?
It combines meaning-based search (pgvector) with keyword search (BM25), re-ranks the results, answers with numbered citations and says it couldn't find something instead of guessing.
Can I measure the quality of AI answers?
Yes. The answer quality lab scores test questions through the real answer path against a random baseline, reports latency and cost, runs safety tests and proposes tuning changes for you to approve.