Dograh

Quick Verdict

Dograh is a BSD 2-Clause voice-agent platform built around visual call flows, replaceable speech and language providers, telephony integrations, and self-hosting. As of July 20, 2026, its repository remains active, with release v1.42.0 published in July, and the company offers the same product in open-source, metered cloud, and private enterprise forms. It is therefore a current platform, not an abandoned demo.

The strongest reason to choose Dograh is control: a team can inspect the source, define explicit workflow branches, bring provider keys, run inside its own infrastructure, and test calls before connecting production numbers. That control comes with responsibility. “Free self-hosted” does not mean free calling; servers, storage, observability, phone numbers, carriers, speech recognition, language models, text-to-speech, and operations all add cost. A deployment also needs lawful contact data, AI identity disclosure, applicable recording consent, opt-out handling, calling-hour controls, human escalation, and fraud safeguards.

Best For

• Engineering teams building structured inbound support, appointment, qualification, or internal-service call flows.
• Organizations that require source review, provider portability, private deployment, or data-residency control.
• Product teams that want to test voice and text turns in a browser before paying for production telephony.
• Companies that already operate CRM, monitoring, incident response, security, and human support processes.
• Not suitable for anonymous mass calling, impersonation, deceptive payment collection, harassment, or teams without legal and operational ownership.

Key Features

• Visual workflow builder: Start, agent, global instruction, tool, transition, and end-call elements make branching behavior easier to inspect than one oversized prompt.
• Browser test modes: Test Audio exercises real voice turns, while Test Chat supports faster iteration, replay, and transition review without a phone number.
• Telephony and handoff: The official repository lists Twilio, Vonage, Telnyx, Plivo, Vobiz, Cloudonix, and Asterisk ARI. Transfer support depends on the selected provider and region.
• Replaceable voice stack: Teams can use an STT-LLM-TTS cascade or supported speech-to-speech models, bring their own provider credentials, and connect locally served components where practical.
• Production integration: Knowledge bases, webhooks, tool calls, call variables, object storage, and Python and Node SDKs can connect calls to CRM, scheduling, or ticketing systems.
• MCP and deployment choices: Compatible coding agents can inspect documentation and node schemas and save workflow drafts. Docker self-hosting, managed cloud, and operated private deployment carry different cost and support boundaries.

Use Cases

A defensible deployment starts with a narrow, disclosed job: route an inbound support caller, confirm an appointment for a customer who requested reminders, collect non-sensitive qualification data from an opted-in lead, or provide an internal IT help line. Build AI identification, purpose explanation, recording notice, consent capture where required, data minimization, opt-out, human transfer, and emergency handling into explicit workflow nodes. Test silence, interruption, accents, background noise, tool timeouts, invalid data, and transfer failure before opening production traffic.

Dograh should not be used to imitate a bank, government body, employer, celebrity, or known individual; hide that a caller is synthetic; repeatedly call a person who opted out; or request passwords, one-time codes, or irreversible payments. High-impact medical, legal, credit, employment, and account-recovery decisions require qualified human review. Synthetic or cloned voices should come only from documented, purpose-specific permission, and the recipient should receive a clear synthetic-media disclosure. Minors require especially restrictive defaults, verified guardian authorization where lawful and necessary, and no behavioral marketing or identity simulation.

Pricing

The published cent-per-minute figure covers the Dograh platform layer, not an all-in call. Model the complete cost from average call duration, carrier route, retries, recording retention, chosen STT/LLM/TTS stack, concurrency, and human transfers. A self-hosted installation can eliminate a platform fee while increasing fixed engineering and reliability costs.

Pros

• Auditable BSD-licensed source with a full self-hosted platform rather than a deliberately limited trial.
• Explicit workflow branches make tools, transfers, and failure behavior easier to review.
• Provider portability across telephony, speech, models, and storage reduces one form of lock-in.
• Audio and text test modes expose flow errors before production phone costs begin.
• A managed path allows a quick pilot, while private deployment remains available for mature teams.
• Current SDK and MCP surfaces support programmatic and developer-assisted workflow management.

Cons

• Self-hosting requires container, database, storage, monitoring, upgrade, backup, and incident-response skills.
• The advertised platform rate excludes most of the components that determine the final per-minute bill.
• Every added provider creates another latency, outage, billing, privacy, and contract dependency.
• Real-time speech needs extensive testing for interruption, noise, long pauses, names, and tool delays.
• Private infrastructure does not automatically create HIPAA, GDPR, SOC 2, or telecom compliance.
• Outbound calling can create serious consent, consumer-protection, and reputation risk if governance is weak.

Alternatives

FAQ

Is Dograh actually free?

The self-hosted platform is free under BSD 2-Clause, but operating it is not. Budget for infrastructure, carriers, numbers, speech and model APIs, storage, monitoring, and engineering. Managed cloud adds a platform minute fee.

Can I test without connecting a phone provider?

Yes. Test Audio and Test Chat are useful for workflow iteration. A production launch still requires controlled tests of numbering, transfer, recording disclosure, provider failure, latency, and opt-out behavior.

Does self-hosting make a deployment compliant automatically?

No. It can improve data-location control, but the operator remains responsible for access policy, retention and deletion, contracts, security controls, user rights, calling rules, and evidence that consent was collected.

May I clone an employee’s or public figure’s voice for calls?

Only use a voice with explicit, documented permission covering purpose, term, territory, and withdrawal. Do not imitate public figures, customers, coworkers, or officials to make recipients believe the person is speaking. Clearly disclose synthesis and provide a complaint and takedown route.

How should a team reduce fraud and harassment risk?

Use contacts with a lawful source and defined purpose, enforce calling hours and do-not-call requests, rate-limit retries, block requests for passwords and one-time codes, restrict payment actions, log important events, and route high-risk cases to trained humans.

Can Dograh be used with minors?

Avoid direct deployment to minors by default. Where a legitimate service requires it, obtain valid guardian and organizational approval, minimize collection, prohibit cloned identities and behavioral marketing, add human supervision, and offer deletion and complaint mechanisms.

Bottom Line

Dograh is compelling when ownership of the workflow and deployment boundary matters more than turnkey simplicity. Start in browser test modes, then use test numbers to validate disclosure, consent, transfer, opt-out, failure, and security controls. Compare managed and self-hosted options using the complete call cost, not the platform fee alone. Teams with mature engineering and governance can gain meaningful portability; teams without legal review, quality assurance, and operational coverage should choose a more managed service or postpone automation.