Speechmatics
Also known as: Speechmatics Flow, Flow, Speechmatics API
Speech to text and text to speech layer that voice agents run on, plus Flow, its own voice agent platform, with cloud, on premise, on device and VPC deployment and SOC 2 Type II certification.
Speechmatics supplies the speech layer that voice agents run on, and with Flow it also ships an agent platform of its own. Flow is positioned as an AI voice agents automation platform that triages calls with intelligent routing, proactive call handling and round the clock coverage, built on Speechmatics hosted infrastructure so voice data never has to reach a third party cloud.
The underlying stack is real time and batch speech to text across 56 or more languages with speaker diarization and multilingual code switching, plus text to speech at sub 150 millisecond latency. It integrates into the agent framework ecosystem rather than competing with it, with named support for Pipecat and LiveKit.
Deployment is the standout. Speechmatics runs in the cloud, on premise, on device, on edge hardware, inside a customer virtual private cloud or in hybrid configurations, with globally distributed data processing environments for sovereignty obligations. On device transcription is described as efficient enough for a laptop. For organisations that cannot send audio anywhere, this is among the most complete answers in the index.
Security matches it: SOC 2 Type II certified, ISO/IEC 27001:2022 accredited, GDPR and HIPAA compliant, with a Trust Centre carrying the DPA and a sub business associate agreement. Real time audio is never stored and batch data auto deletes after seven days or sooner via API, and customer data is not used to improve models.
The honest limits are the agent layer above the speech layer. No agent memory, prebuilt agent library or computer use capability is documented, and orchestration is call centric rather than general purpose.
Vendor details
Canonical URL
https://www.speechmatics.com
Category
Agent infrastructure
Subcategory
Speech infrastructure with a voice agent platform
Funding status
Private, UK based. Funding detail not retrieved in this pass.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Developer APIs and SDKs with a platform agnostic build once deploy everywhere posture. Named integrations into the agent framework ecosystem including Pipecat and LiveKit, the latter bringing speaker diarization so agents understand not just what was said but who said it. Custom dictionaries and formatting rules tune the models to domain vocabulary without retraining.
In practice
A team building on Pipecat or LiveKit swaps in Speechmatics for speaker diarization so its agent knows who is speaking in a multi party call, not just what was said.
A regulated organisation runs transcription entirely on premise or on device so audio never leaves its own infrastructure, while still using a current commercial model.
A contact centre uses Flow to triage and route inbound calls automatically, escalating to human agents only where the routing logic cannot resolve the request.
Sources & related URLs
Agentic Index coverage score
8.5 / 14 capabilities · 61%
| Integrations & Tool CallingDeveloper ready APIs and SDKs with a platform agnostic build once deploy everywhere design, plus named integrations into the agent framework ecosystem including Pipecat and LiveKit, the latter adding speaker diarization to LiveKit agents. Speechmatics AI voice agents page and developer documentation 2026-08-07 | Full |
|---|---|
| Workflow OrchestrationFlow is an AI voice agents automation platform that automates and triages calls with intelligent call routing, proactive call handling and round the clock coverage, and Speechmatics ships components for building voice agents on top. Orchestration is call centric rather than general purpose. Speechmatics Flow and AI voice agents pages 2026-08-07 | Full |
| Knowledge Grounding & RAGCustom dictionaries, custom vocabularies and formatting rules tune recognition to domain specific content including medical and legal terminology without retraining, but that is vocabulary adaptation rather than a retrieval, vector or knowledge grounding layer, none of which was retrieved. Speechmatics enterprise and pricing pages 2026-08-07 | Partial |
| Human Oversight & GuardrailsFlow routes and triages calls, which implies escalation to human agents as the fallback path, and Speechmatics publishes guidance on human oversight and agent disclosure obligations, but no approval gate, confidence threshold or documented escalation mechanism for its own agents was retrieved. Speechmatics Flow page and compliance guidance 2026-08-07 | Partial |
| Security, Identity & GovernanceSOC 2 Type II CERTIFIED, ISO/IEC 27001:2022 accredited, GDPR and HIPAA compliant, with a Trust Centre carrying the DPA and a sub business associate agreement; encryption in transit and at rest, and processing on Speechmatics hosted infrastructure so voice data need not reach third party clouds. Real time audio is never stored and customer data is not used to improve models unless the customer opts into the model training discount programme. Speechmatics security and pricing pages 2026-08-07 | Full |
| Observability & AuditabilityProvides detailed insight into speech patterns, confidence scores and conversation analytics, which supports quality monitoring, but no agent execution audit log, trace view or governance reporting surface was retrieved. Speechmatics product pages 2026-08-07 | Partial |
| Memory & State PersistenceNo agent memory, conversation state store or cross session persistence model was documented. The data posture runs the other way: real time audio is never stored and batch data auto deletes after seven days or sooner. Honest absence. Speechmatics pricing and security pages 2026-08-07 | Unable to verify |
| Deployment & Data ResidencyAmong the most complete in the index: cloud, on premise, ON DEVICE and edge, customer VPC and hybrid, with globally distributed data processing environments explicitly for compliance, regulatory and sovereignty obligations. On device transcription is stated to be efficient enough to run on a laptop. Speechmatics enterprise and security pages 2026-08-07 | Full |
| Prebuilt Agents, Templates & PacksPublished use case guidance exists for call automation, education and commerce, but no library of prebuilt agents, templates or starter packs was documented as a product surface. Honest absence. Speechmatics Flow and use case pages 2026-08-07 | Unable to verify |
| Triggers & Channel CoverageReal time streaming and batch processing across telephony and audio channels, scaling to thousands of concurrent audio streams, with Flow handling inbound calls continuously and sub second speech to text plus sub 150 millisecond text to speech. Speechmatics enterprise and Flow pages 2026-08-07 | Full |
| Model Flexibility & RoutingSpeechmatics supplies its own speech to text and text to speech models and states a platform agnostic build once deploy everywhere posture with models runnable in cloud, on premise or on device, but no large language model selection, provider choice or routing configuration for the agent layer was retrieved. Speechmatics enterprise and Flow pages 2026-08-07 | Partial |
| APIs, SDKs & MCP ExtensibilityDeveloper ready APIs and SDKs documented for fast integration, a flexible architecture customisable to enterprise needs, custom vocabulary and formatting configuration, and first class integration into third party agent frameworks Pipecat and LiveKit. No MCP support was retrieved. Speechmatics developer and AI information pages 2026-08-07 | Full |
| Testing, Debugging & OptimizationSpeechmatics documents an integration giving agent developers a QA layer for real world complexity and publishes guidance on voice agent testing frameworks, but no first party evaluation harness, scoring or regression suite for customer built agents was retrieved. Speechmatics AI voice agents resources 2026-08-07 | Partial |
| Browser & Computer UseSpeechmatics operates on audio streams and telephony; no browser control, page navigation or computer use capability is documented anywhere on the product surface. Speechmatics product documentation 2026-08-07 | Unable to verify |
The Agentic Index coverage score grades every vendor Full, Partial or Unable to verify against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Usage based; 100 USD free credit to start, no card required
usage based on audio processed, with volume tiers
Included quota
100 USD of starting credit. Realtime audio is never stored and batch data auto deletes after seven days or sooner via the API for customers not in the training programme.
What is public
Plan structure, the free starting credit, usage based model, deployment flexibility and the existence of the model training discount programme are all published, along with the full certification list. Specific unit rates exist on the page but were not captured here.
Billing mechanics
Self serve signup with starting credit and no card, moving to usage based billing. Discounted pricing scales with volume. Enterprise plans are quoted for unlimited scale and flexible deployments, with a dedicated Customer Success Manager and Solutions Engineer.
Cost watchouts
The headline rate depends on whether the customer has opted into the model training discount programme, so a quoted price may carry a data licensing condition attached. On premise, on device and VPC deployments shift infrastructure and operational cost onto the customer. Per minute or per hour rates by product line and language were not captured in this pass, so the published page should be read directly before modelling volume.
Variable cost rationale
Usage based billing on audio volume is inherently variable, but it is measured in a unit a buyer can forecast directly from call minutes or file hours, unlike the opaque agent consumption units used across the BPM and agent builder pockets. Graded medium rather than high because the meter is legible and a free credit allows real measurement before commitment, with the caveat that the effective rate depends on the model training discount election.
Additional watchouts
Confirm which side of the model training discount programme a quote assumes before comparing rates against another vendor. The cheaper number may carry a data licensing condition.
Overage / add-ons
Usage based, so consumption bills as incurred against credit rather than triggering plan overage.
Sales call required
Mixed (some tiers require a call)
Free / trial
100 USD in credit on signup, no payment card required
Lowest paid plan
Usage based self serve
Commercial notes
**DATA FOR DISCOUNT IS A COMMERCIAL MECHANISM NOT SEEN ELSEWHERE IN THIS INDEX: the model training discount programme prices customers on whether they permit training on their speech data. Every other vendor treats no training on customer data as a flat assurance; here it is a purchasable variable.** Worth asking of other vendors: is data retention priced?
Key ambiguities
The specific per unit rates by product line, and the size of the model training discount. Both are on the public page but were not captured as figures in this pass, so the bucket here reflects the free entry point rather than a modelled workload.
Missing data
Per unit rates by product line and language, and the magnitude of the model training discount.
Related vendors
- Acrab — Singapore compute infrastructure company building a full stack…
- AgentOps — Agent observability and reliability platform with broad model and…
- Agno — High-performance agent runtime and framework (formerly Phidata) with…
- AIsa — Unified resource and payment gateway for AI agents that lets them…
- AlphaBitCore — AI control plane that governs how models, agents, tools, and…
- Anchor Browser — Cloud hosted browser infrastructure that lets AI agents operate real…
Alternatives to Speechmatics
The closest documented capability profiles to Speechmatics among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Gentoro6.5 / 14A lighter documented profile than Speechmatics
- Arcade8.0 / 14Adds documented Prebuilt Agents, Templates & Packs
- Deepgram6.0 / 14A lighter documented profile than Speechmatics
- Pipecat10.0 / 14Adds documented Memory & State Persistence and Prebuilt Agents, Templates & Packs
- Thread AI10.0 / 14Adds documented Memory & State Persistence and Prebuilt Agents, Templates & Packs
- Windmill10.0 / 14Adds documented Memory & State Persistence and Prebuilt Agents, Templates & Packs
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded