Dograh

Based on 32 comments (primarily HackerNews with some Reddit), Dograh generates strong enthusiasm in the developer community. The dominant motivation is cost: one user reports that with Vapi, 60-70% of total spend was the platform fee while only 30-40% covered actual LLM/STT/TTS usage — 'platform cost dominated everything.' Users appreciate that Dograh extends pipeline functionality with custom Frames and Processors, addressing the 'plumbing' pain that exists even with good OSS stacks like Pipecat and LiveKit ('no standard way to extract variables from conversations'). The platform serves use cases from large BPO call centers (outbound and inbound) to individual builders. Latency is heavily discussed: creators transparently report 500-600ms with colocated/self-hosted models, 800-1000ms with Gemini 2.5 Flash, and higher with larger models like GPT-4.1. Users acknowledge the fundamental latency-vs-reasoning tradeoff — smaller colocated models are faster but worse at instruction following. Feature gaps noted: no voice personas selector yet (like Vapi has), which creators confirm is 'in the pipeline.' Deeper architectural concerns raised: one user warned against shipping secrets in configs/prompts and recommended server-side credential injection with audit trails. Another questioned how voice-agent-as-MCP handles unreliable return values when chaining to CRM/email MCPs, noting voice calls produce messy transcripts and ambiguous dispositions that break downstream typed fields. Community sentiment is overwhelmingly positive and supportive (project hit its 100th GitHub star during the discussion), though some users found the original HN title misleading — sounded like Vapi was open-sourced, not that someone built an alternative. The creators clarify: 'Dograh is an open source alternative to Vapi, not a clone though. Like saying n8n is a clone of Zapier because they solve the same problem.'

No aggregate ratings were found for this product during the last harvest.