Knowledge graph

Ask your meetings anything — without OpenAI

Semantic search across your entire AudioMap library with on-prem embeddings (bge-m3 in Europe). Ask "what did client X say about pricing?" and get answers with exact note + timestamp citations.

Try for free See pricing

🇪🇺

Embeddings on-prem in Europe

bge-m3 (1024 dims) on Hetzner DE. Your transcripts never reach OpenAI. Vector search via pgvector on the same Postgres.

🔍

Cross-library questions

Ask "which clients mentioned a price increase?" across all your notes. AudioMap finds them with citations.

📍

Timestamp citations

Every answer cites note + exact second. Click and AudioMap jumps to that segment. No "trust the AI" — verify what was said.

🎯

Per-note chat

Or scope your question to a single recording. Useful for asesorías reviewing a specific client meeting.

🌐

MCP server

Connect your AudioMap library to Claude, ChatGPT, Cursor or Gemini. Use them as your memory.

⚡

Fast even at scale

pgvector + cosine similarity scales to thousands of meetings without external vector DB. Sub-200ms typical query latency.

Feature	AudioMap	Otter / Fireflies / Fathom
Cross-library semantic search	✓ via embeddings	Partial — keyword only
Embeddings provider	✓ bge-m3 on-prem EU	✗ OpenAI USA
Timestamp citations	✓ click-to-jump	✗
MCP server	✓ for Claude/ChatGPT/Cursor	Partial
Vector DB	✓ pgvector (same Postgres)	External (Pinecone/Weaviate)
Data residency	✓ Hetzner DE	✗ USA

Why bge-m3 instead of OpenAI text-embedding-3?+

bge-m3 (BAAI) is multilingual (100+ languages including all Spanish variants), open-source, and runs entirely on our Hetzner DE server. Quality is comparable to text-embedding-3-large for Spanish. We avoid OpenAI specifically to keep all data in Europe.

How big can my library be?+

pgvector handles millions of segments efficiently with HNSW index. Practical limit is your disk, not the engine. We've tested with 500K+ segments in single-digit-second query latency.

Can I ask questions in English about Spanish recordings?+

Yes. bge-m3 is cross-lingual: the embedding for "what did Sara say about budget?" in English matches Spanish content discussing "presupuesto". Cross-language retrieval works both directions.

How does MCP work?+

Generate an API key at /dashboard/api-keys, configure your Claude/ChatGPT/Cursor with our MCP server URL, and they can query your AudioMap library directly. Tools: search_notes, get_transcript_segments, ask_note, etc. Free tier includes 50 credits/month MCP usage.

Turn your library into your memory

Semantic search included from Free tier. 120 free minutes monthly. Data stays in Europe.

Try for free