Dial

Dial is communications infrastructure built for AI agents. One number can place and receive calls, exchange SMS, and use iMessage with RCS/SMS fallback. Integration options include REST, Node.js and Python SDKs, LangChain, a signed unified webhook, and a first-party MCP server. The vendor lists Claude Code, Cursor, Windsurf, Grok, and other MCP clients.

Voice supports inbound/outbound calls, real-time transcription, and TTS. Self-hosted mode uses the customer's LLM over WebSocket; managed mode supplies the end-to-end model. Real US, Canadian, and international numbers are provisioned through the API. Compare primitives from Twilio, agents from Bland AI and Voiceflow, and speech from ElevenLabs.

Pay-as-you-go has no base fee or minimum. A standard number is $3/month. Voice starts at $0.13/minute self-hosted or $0.22/minute managed. US SMS is $0.02/message; international tiers range from $0.02 to $1.00/message. An iMessage number is $250/month, with iMessage traffic listed free. New accounts receive $5 credit without a card. Enterprise is custom under an annual agreement, adding volume rates, onboarding, an account manager, SLA, and custom invoicing. 10DLC/A2P registration carries an additional fee and an advertised three-to-five-business-day wait.

At 10,000 managed minutes, voice alone starts near $2,200 before numbers, messages, context tools, tax, or other services. Customers remain responsible for recording consent, opt-outs, quiet hours, campaign registration, and authorization. Start with allowlisted recipients, spending caps, signed-webhook verification, and approval before purchases or bookings.

Practical evaluation and verdict

Run a two-week pilot on one bounded, production-like workflow with at least 50 representative tasks and a human-reviewed baseline. Record completion rate, factual or technical errors, p50 and p95 latency, total usage cost, setup time, correction time, and failure categories. Include permission failures, stale inputs, ambiguous requests, service outages, and rollback cases. Start with read-only access or a sandbox. Add write actions only after an owner defines approval and recovery steps.

Review retention, deletion, data export, subprocessors, regional processing, rate limits, authentication, audit logs, and whether customer content is used for training. Calculate annual cost at realistic volume with a 25% usage buffer; include external model/API fees, implementation, monitoring, and reviewer labor. Public prices can change, so confirm checkout or contract terms before purchase.

The best fit is a team with a measurable bottleneck that matches the features above. It is a weak fit when nobody owns output review, permissions are broader than the task, or a deterministic script would be safer and cheaper. A successful pilot should save net time after correction and governance—not merely produce an impressive demo. Keep a fallback process, cap spending where usage is variable, and periodically rerun the same evaluation set after product or model changes.