Top Builders

Explore the top contributors showcasing the highest number of app submissions within our community.

Speechmatics Flow

Speechmatics Flow is a speech-to-speech API for building real-time conversational AI agents. Announced in July 2024, it combines Speechmatics' speech recognition with an LLM and text-to-speech into a single API connection, removing the need to stitch together separate transcription, inference, and synthesis services. Flow handles the real-time challenges of two-way voice conversations, including turn detection, interruption management, and multi-speaker isolation.

General
AnnouncedJuly 30, 2024
DeveloperSpeechmatics
TypeVoice agent API (speech-to-speech)
LicenseCommercial API
Documentationdocs.speechmatics.com/voice-agents-flow
GitHubspeechmatics/speechmatics-flow

Core Features

  • End-to-end speech-to-speech pipeline: STT, LLM, and TTS via a single API call.
  • Smart turn detection: uses a small language model (SLM) to decide when a speaker's turn has ended, reducing false triggers.
  • Interruption handling: ignores unintentional interruptions, handles intentional ones gracefully.
  • Speaker locking: isolates a target speaker and filters out background voices in multi-speaker environments.
  • Function calling: connects agents to external tools, APIs, databases, and validation services.
  • Internet search: agents can query live web data (weather, news) during conversations.
  • 55+ language support: same multilingual coverage as Speechmatics' STT API.
  • Conversation moderation: real-time transcript analysis to flag or filter content.
  • Flexible deployment: private SaaS cloud or on-premises.
  • Security: ISO/IEC 27001:2022 certified, GDPR compliant.

Pricing

TierIncluded
FreeUp to 50 hours/month
EnterpriseCustom pricing, contact Speechmatics

Tools and Resources

Ecosystem and Integrations

  • Works with Vapi for no-code voice agent deployments.
  • Works with LiveKit for WebRTC-based real-time infrastructure.
  • Works with Pipecat for open-source voice pipeline orchestration.
  • Supports contact center, healthcare, drive-thru, educational assistant, and smart device use cases.

Get started with the free tier (50 hours/month) or contact Speechmatics for enterprise access at flow-help@speechmatics.com.

Edit on GitHub

speechmatics Speechmatics flow AI technology Hackathon projects

Discover innovative solutions crafted with speechmatics Speechmatics flow AI technology, developed by our community members during our engaging hackathons.

Dementia ASR Screening

Dementia ASR Screening

Overview The Dementia ASR Screening & Multi-Factorial Risk Stratification Application is an advanced digital health portal designed to bridge the gap between AI-driven speech biomarkers and modifiable clinical risk factors. Built on a premium, mobile-first dark theme interface, the portal enables patients to undergo non-invasive, visual speech elicitation tests in seconds, generating comprehensive, weighted dementia conversion risk assessments. Key Features & Technological Innovation Lossless Acoustic Elicitation: Features a custom-built, client-side Web Audio API recorder that encodes raw microphone samples directly into lossless 16-bit Mono PCM WAV format. This ensures 100% processing compatibility with cloud-based Automatic Speech Recognition engines. Speechmatics SaaS ASR Integration: Orchestrates advanced speech analyses via Express backend proxies. By disabling disfluency filtering (remove_disfluencies: false), the portal natively captures and tags spoken filler words. Acoustic Pause Biomarker Extraction: Reads word-level transcript timestamps to automatically isolate and flag long conversational hesitation gaps exceeding 1.5 seconds, a key marker in acoustic cognitive screens. Wiley 2025 Multi-Factorial Stratification: Incorporates a validated clinical risk algorithm based on the October 2025 study in Alzheimer's & Dementia (DOI: 10.1002/alz.70870): Age Scoring, Cognitive Reserve, BMI,Depression Screening (PHQ-2), Weighted Clinical Risk Dial: Automatically weighs Vocal speech biomarkers (40% weight) against Clinical modifiable factors (60% weight) to render a composite, color-coded Stratified Risk Indicator (Low, Moderate, or High Risk). Comprehensive Session Review Timeline: Connects to a self-healing SQLite database to store serialized disfluency details, allowing patients to inspect past tests and review historical reports exactly as they looked live along with an Interactive Demographics Editor.

Executor

Executor

Executor is an autonomous meeting-to-execution enforcement system built for enterprise teams that lose accountability after meetings. Most AI meeting tools stop at summaries, but Executor continues into execution by extracting commitments, identifying owners, deadlines, priorities, and source quotes, then turning those decisions into tracked work. Users can upload transcripts or run live meetings with Speechmatics real-time transcription. Gemini analyzes the meeting, extracts structured action items, scores execution risk, and powers follow-up drafting. Each commitment can be synced directly into Jira, Asana, and Notion, connecting meeting decisions to the tools teams already use. Executor also runs a background enforcement agent deployed on Vultr. The monitor checks deadlines every six hours, marks overdue commitments, drafts follow-up emails, retries failed sends, and escalates to managers after repeated missed follow-ups. Human review remains available before sensitive messages are sent, while every autonomous action is stored in an audit log. The application is deployed publicly on Cloudflare Workers, uses Supabase for authentication and database storage, Gemini for reasoning, Speechmatics for voice transcription, and Vultr for the scheduled enforcement worker. Executor closes the gap between “we decided” and “it actually got done.”

VoxCall Oracle: Live Audio Trading Agent

VoxCall Oracle: Live Audio Trading Agent

The latency gap in modern finance is a multi-million dollar problem. Traditional algorithmic trading relies on text-based transcripts that humans or bots read minutes after words are actually spoken on an earnings call. By the time the data is digested, the market has already moved. VoxCall Oracle is an end-to-end, production-ready pipeline designed to bridge the gap between raw audio and instant market execution. Our fully autonomous agent listens to live earnings calls and executes trades in real-time without human intervention. We utilized four core technologies to build this stack: 1. Speechmatics: We use their best-in-class speech-to-text API for real-time transcription and speaker diarization. This allows our agent to know exactly *what* was said and *who* said it (e.g., separating a confident CEO from a cautious reporter). 2. Featherless AI: To eliminate the risk of AI hallucinations, we route the transcript chunks into a 3-model financial ensemble via Featherless (Llama-Open-Finance, Fin-o1, and finance-chat). These specialized models vote on the market sentiment to ensure pinpoint accuracy. 3. LangGraph: In automated trading, safety is critical. We built a LangGraph state-machine that acts as a risk-gating firewall. Trades are only approved if the AI ensemble's confidence score exceeds a strict user-defined threshold (e.g., >75%). 4. Kraken CLI: Approved signals are instantly routed to the Kraken API for ultra-low-latency order execution. Our entire platform is wrapped in a premium, glassmorphism UI deployed on Streamlit Cloud, featuring a live Paper PnL tracker and execution logs. VoxCall Oracle isn't just a script—it's a fully functional, cloud-native hedge fund analyst.

Close loop health

Close loop health

B2B SaaS, services, and subscription companies lose 20-40% of annual revenue to churn that was preventable. The brutal part: in post-mortems, the warning signs were always there — a champion stopped replying in week 3, support tickets got terser in week 5, the QBR got rescheduled twice in week 7. By the time Customer Success heard "we're not renewing" in week 11, the internal decision had been made weeks earlier. The reason this keeps happening isn't laziness. A CSM managing 40-80 accounts cannot read every Slack-shared email thread, listen to every support call, watch every usage dashboard, and notice that procurement at one account just started LinkedIn-stalking a competitor. The signals exist; the human bandwidth doesn't. ClosedLoop is the agent system that watches everything, reasons about what it sees, and either acts or escalates with the save-play already drafted.

AgentOps 360

AgentOps 360

AgentOps 360 is an autonomous enterprise operations agent that transforms unstructured business input into structured execution plans. In many companies, meetings create a lot of information, but decisions, tasks, risks, owners, and follow-up messages are still handled manually. AgentOps 360 solves this by turning meeting notes, uploaded documents, PDFs, and audio transcripts into a clear business dashboard. The system uses a multi-agent workflow with five specialized agents: an Intake Agent, Reasoning Agent, Planning Agent, Communication Agent, and Audit Agent. Together, they analyze the input, extract decisions, generate action tasks with owners and priorities, detect risks with mitigation strategies, draft professional follow-up emails, and create an auditable agent trace. AgentOps 360 is powered by Gemini for reasoning and structured analysis, Speechmatics for audio transcription, PdfPig for PDF text extraction, React and TypeScript for the frontend, and ASP.NET Core with SQL Server for the backend and persistence. Each analysis is saved in the database, allowing users to reopen previous runs, track execution history, and export a structured executive report. The app also includes a fallback engine to keep the workflow reliable if the AI provider is temporarily unavailable. AgentOps 360 is designed for managers, founders, operations teams, HR teams, sales teams, and customer support teams who need to move faster from conversation to execution.