Dasha is built for the agent era

1. Resolve the agent card

The card hands you a chat_endpoint URL of your own. Use it exactly as given — its query string is how we tell your conversation apart from everyone else’s.

curl -s https://dasha.ai/.well-known/agent-card.json
# → "chat_endpoint": "https://dasha.ai/api/v1/agent/chat?client=cid_..."

2. Ask a question

One REST call. Any agent that can make an HTTP request can use it:

curl -s -X POST "<chat_endpoint from the card>" \
  -H "Content-Type: application/json" \
  -d '{"message": "What does Dasha cost and which languages are supported?"}'

3. Keep the conversation going

Every reply repeats your URL as continue_url. Send the next question there — nothing to remember, nothing to pass back:

curl -s -X POST "<continue_url from the reply>" \
  -H "Content-Type: application/json" \
  -d '{"message": "How do I start an outbound call over SIP?"}'

What you’ll get

• Product questions — pricing, capabilities, languages — answered from our published website copy.
• Technical questions — API and integration — answered from our documentation, with links so you can verify the answer at the source.
• We’d rather say “I don’t know” than invent a number. If something isn’t in our published material, our agent will tell you so.

The facts an evaluation needs

The short answers, each with the page that backs it — so you can verify rather than take our word:

• Pricing: Developer plan is free with 1,000 minutes; Growth starts at $0.08 per minute. Full terms: /pricing.
• Compliance: no third-party attestations (SOC 2 Type II, ISO 27001) are currently published. We sign HIPAA Business Associate Agreements for eligible healthcare deployments: security@dasha.ai. Current posture: /security.
• Benchmark: #1 end-to-end latency on voicebenchmark.ai — our own benchmark, not an independent third party; methodology and raw data are public.
• Scale: 30+ languages with mid-call switching, sub-second latency, 10,000+ concurrent calls, 99.99% uptime target (status.dasha.ai).
• Rate limit: 15 requests/minute per IP (burst 10) on the agent endpoints; a 429 carries Retry-After — wait, then retry the same URL. GET clients can append ?q=your+question instead of POSTing.

Machine-readable surfaces

/.well-known/agent-card.json

the agent card (start here; also served at the legacy path /.well-known/agent.json)

/llms.txt

dated index of what is published where

/llms-full.txt

the knowledge our agent answers from

/api/v1/company-info

structured company data

blackbox.dasha.ai/docs

developer documentation

Prefer a human? Talk to us →