The most important context for AI isn't written down.
It's spoken. Recall.ai is the API that turns human conversation into data AI can use. We're the infrastructure layer thousands of AI products are built on.
$ curl -X POST https://api.recall.ai/api/v1/bot/ \
-H "Authorization: Token $RECALL_API_KEY" \
-d '{ "meeting_url": "https://zoom.us/j/9834923" }'
{ "id": "bot_7f3a91", "status": "joining_call" } Hyperscale infrastructure to process millions of concurrent media streams
We are at the earliest stages of our mission, but are already running at a scale most companies never reach: a cluster of 100,000 machines, processing 5 terabytes of data per second.
Last year, that covered 0.02% of the world's office worker conversations. The other 99.98% is next.
A Black-Friday scale traffic spike, 240 times per week
Every meeting in the world starts at the top of the hour or the half. Millions of people click "start" within seconds of each other — a Black-Friday-scale surge, 48 times a day.
The bottleneck on AI is context, not intelligence
LLMs are super-human at producing work, but they don't know what humans know: the nuances of the customer's use-case, the reasoning behind each line of the spec, or that a similar project was attempted six months ago.
Context starts as conversation. Agents can't access any of it.
In every company, humans get most of their context through conversations, which agents can't access. Agents only access the small, compressed sliver that gets written down in messages, documents, and emails.
A single year of US office conversations contain 500% more words than the entire internet.
Recall is the API to record, understand and search conversations as context for AI
The raw torrent of audio, video, and metadata gets packaged into structured context, archived alongside the raw data, and retrieved in milliseconds — one query away from any agent.
We write about the hard parts.
Segfaults, serial ports, and the many ways Postgres does not scale — the problems you'd actually work on here.
Connect AI to the world's context.
We're a small team in San Francisco solving hard infrastructure problems with unusual leverage. If that sounds like your kind of work, we'd like to talk.
The most important context for AI isn't written down.
It's spoken. Recall.ai is the API that turns human conversation into data AI agents can use. We're the infrastructure layer thousands of AI products are built on.
$ curl -X POST https://api.recall.ai/api/v1/bot/ \
-H "Authorization: Token $RECALL_API_KEY" \
-d '{ "meeting_url": "https://zoom.us/j/9834923" }'
{ "id": "bot_7f3a91", "status": "joining_call" } Hyperscale infrastructure to process millions of concurrent media streams
We are at the earliest stages of our mission, but are already running at a scale most companies never reach: a cluster of 100,000 machines, processing 5 terabytes of data per second.
Last year, that covered 0.02% of the world's office worker conversations. The other 99.98% is next.
A Black-Friday scale traffic spike, 240 times per week
Every meeting in the world starts at the top of the hour or the half. Millions of people click "start" within seconds of each other — a Black-Friday-scale surge, 48 times a day.
The bottleneck on AI is context, not intelligence
LLMs are super-human at producing work, but they don't know what humans know: the nuances of the customer's use-case, the reasoning behind each line of the spec, or that a similar project was attempted six months ago.
Context starts as conversation. Agents can't access any of it.
In every company, humans get most of their context through conversations, which agents can't access. Agents only access the small, compressed sliver that gets written down in messages, documents, and emails.
A single year of US office conversations contain 500% more words than the entire internet.
Recall is the API to record, understand, store, and search conversations.
The raw torrent of audio, video, and metadata gets packaged into structured context, archived alongside the raw data, and retrieved in milliseconds — one query away from any agent.
We write about the hard parts.
Segfaults, serial ports, and the many ways Postgres does not scale — the problems you'd actually work on here.
Connect AI to the world's context.
We're a small team in San Francisco solving hard infrastructure problems with unusual leverage. If that sounds like your kind of work, we'd like to talk.
