ROLE
Voice platform
The call loop: listen, think, speak.
81
73–89 · low confidence
STACKFOR
StackFor building a phone or voice agent
Latency is part of the job. A beautiful narrator that answers a second late is the wrong pick for this role.
A voice that can hold a phone call: a voice platform, a speaking voice, speech-to-text, a model, a phone number, and a way to log the result.
Agent fanBuilderAdvancedProvisional scores
STACK SCORE
85
Mean of the role winners, compatibility +3. Separate from each tool score.
Provisional seed scores. Method seed-v0-provisional, confidence low, not hands-on tested. The band around a number is a ±8 display label, not a measured interval. Ordering uses the point estimate.
We may earn a commission if a tracking link is connected and you buy through it. It never changes ranks. In this v1 seed, tracking URLs are empty, so “Open site” goes to the vendor. Sponsorship rules.
Recommended stack
Platform, voice, transcription, model, phone number, and a log. Built for a call, not for a YouTube voiceover.
Example compatibility note, not a measured integration test. Vapi and Retell both expect you to bring STT, TTS, and a model.
ROLE
TTS / voice
The speaking voice.
Recommended
ElevenLabsQuality is the reason it leads. Latency still needs a test on your call.
New toolShipped this week
86
78–94 · low confidence
ROLE
Speech to text
The listening side.
82
74–90 · low confidence
Strong transcription. Confirm streaming fit for your call path.
ROLE
LLM API
The model call inside the product.
82
74–90 · low confidence
Short tool-using replies are a known path.
Worth a bake-off. Not the default in most voice samples.
Ranked tools
Sorted by StackFor Score only. A tool can appear twice when it fills two roles. Affiliate programs are not on this list.
- #1
86
78–94 · low confidence
Quality is the reason it leads. Latency still needs a test on your call.
Goal fit25%90Output quality20%94Ease for vibe coders15%86Pricing and free tier15%62Integrations and API10%88Momentum10%90Community sentiment5%84No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open ElevenLabs - #2
83
75–91 · low confidence
Same job, self-hosted if the call data is sensitive.
Goal fit25%80Output quality20%82Ease for vibe coders15%74Pricing and free tier15%88Integrations and API10%90Momentum10%86Community sentiment5%84No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open n8n - #3
82
74–90 · low confidence
Excellent model. Watch latency on a live call.
Goal fit25%84Output quality20%90Ease for vibe coders15%74Pricing and free tier15%66Integrations and API10%86Momentum10%88Community sentiment5%82No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open Anthropic API - #4
82
74–90 · low confidence
The usual streaming ear for this stack.
Goal fit25%90Output quality20%88Ease for vibe coders15%76Pricing and free tier15%68Integrations and API10%86Momentum10%74Community sentiment5%78No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open Deepgram - #5
82
74–90 · low confidence
Short tool-using replies are a known path.
Goal fit25%88Output quality20%88Ease for vibe coders15%76Pricing and free tier15%64Integrations and API10%94Momentum10%86Community sentiment5%74No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open OpenAI API - #6
81
73–89 · low confidence
Worth a bake-off. Not the default in most voice samples.
Goal fit25%82Output quality20%84Ease for vibe coders15%76Pricing and free tier15%78Integrations and API10%84Momentum10%82Community sentiment5%72No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open Gemini API - #7
81
73–89 · low confidence
Realtime stack you can own more of. More assembly.
Goal fit25%84Output quality20%86Ease for vibe coders15%64Pricing and free tier15%88Integrations and API10%82Momentum10%78Community sentiment5%76No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open LiveKit Agents - #8
80
72–88 · low confidence
Strong transcription. Confirm streaming fit for your call path.
Goal fit25%84Output quality20%84Ease for vibe coders15%80Pricing and free tier15%70Integrations and API10%82Momentum10%72Community sentiment5%76No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open AssemblyAI - #9
80
72–88 · low confidence
Aimed at live agents, where delay beats a studio read.
Goal fit25%88Output quality20%86Ease for vibe coders15%74Pricing and free tier15%70Integrations and API10%76Momentum10%82Community sentiment5%70No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open Cartesia - #10
80
72–88 · low confidence
Logs the call into a CRM or a sheet without code.
Goal fit25%82Output quality20%78Ease for vibe coders15%88Pricing and free tier15%72Integrations and API10%94Momentum10%70Community sentiment5%76No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open Make - #11
80
72–88 · low confidence
Developer API for the call loop.
Goal fit25%88Output quality20%82Ease for vibe coders15%80Pricing and free tier15%64Integrations and API10%84Momentum10%80Community sentiment5%72No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open Vapi - #12
79
71–87 · low confidence
Phone-agent product with more of the call center included.
Goal fit25%86Output quality20%80Ease for vibe coders15%82Pricing and free tier15%66Integrations and API10%80Momentum10%76Community sentiment5%74No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open Retell AI - #13
78
70–86 · low confidence
Numbers and call control with the most examples.
Goal fit25%90Output quality20%86Ease for vibe coders15%64Pricing and free tier15%66Integrations and API10%96Momentum10%60Community sentiment5%72No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open Twilio - #14
77
69–85 · low confidence
Open-source pipeline. You host every piece.
Goal fit25%78Output quality20%80Ease for vibe coders15%58Pricing and free tier15%96Integrations and API10%74Momentum10%74Community sentiment5%72No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open Pipecat - #15
77
69–85 · low confidence
Real alternative. Fewer copy-paste voice-agent guides.
Goal fit25%82Output quality20%80Ease for vibe coders15%68Pricing and free tier15%78Integrations and API10%82Momentum10%66Community sentiment5%70No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open Telnyx - #16
75
67–83 · low confidence
A narration studio. The wrong shape for turn-taking.
Goal fit25%72Output quality20%78Ease for vibe coders15%88Pricing and free tier15%68Integrations and API10%74Momentum10%64Community sentiment5%74No cited evidence stored. seed-v0-provisional. Not hands-on tested.
Open Murf
Draft feature checklist
Goal fit is scored against these must-haves. The list is a draft rubric, not a test log.
Voice platform
- Turn-taking
- A path to a phone call
- You can swap the model
TTS / voice
- Low enough latency for a conversation
- An API
Speech to text
- Streaming transcription
- An API
LLM API
- Tool calling
- Short replies are possible
Telephony
- Buy a number
- Webhooks for call events
Automation
- Write the call outcome somewhere a human checks
Questions
Why isn't ElevenLabs the whole stack?
ElevenLabs is a voice. A phone agent also needs something to place the call, something to listen, a model, and a log. Those are separate roles.
Do open-source options get a bonus for being open source?
No bonus for the license. Pricing and ease are already criteria. Pipecat and LiveKit win or lose on those subscores like everything else.
NEWSLETTER
The weekly AI stack drop
New tools worth your time, by use case. One email a week. A sponsor, when there is one, is labeled.