# Softcery

> Softcery builds the conversational AI layer for B2B SaaS platforms: production voice and text agents, end to end, on your infrastructure.

Every page below is authored markdown. Fetch it with `Accept: text/markdown`, or append `.md` to the url (home: `/index.md`).

## Main Pages

- [Softcery | Conversational AI for B2B Software Platforms](https://softcery.com/): Softcery builds the conversational AI layer for B2B SaaS platforms: production voice and text agents, end to end, on your infrastructure.
- [Legal – Softcery](https://softcery.com/legal): Privacy policy and terms of use for the Softcery site and its voice and text agent: what is collected, who processes it, your rights, and the rules of use.
- [Free AI Voice Agent Cost & Latency Calculator 2026](https://softcery.com/ai-voice-agents-calculator): Free 2026 calculator for AI voice agent cost and latency: per-minute pricing across 22 LLMs, 40 STT/TTS models, 12 platforms and 11 transports.
- [AI in Production | Softcery Cases](https://softcery.com/cases): The deployment record: copilots, voice agents, and AI systems Softcery shipped to production, per case.
- [Get in Touch – Softcery](https://softcery.com/contact): Send Softcery an inquiry. Name, email, message; the team reads every wire. Or email hey@softcery.com.
- [AI Receptionist for Hotels | Live Demo | Softcery](https://softcery.com/demos/hotel-receptionist): AI voice receptionist for hotels: answers calls, takes reservations, and handles guest requests around the clock in multiple languages. Recorded demo included.
- [Live Voice AI Demos | Softcery](https://softcery.com/demos): Demonstration voice agents, live on the bench: legal receptionist, hotel receptionist. Call one, it picks up.
- [AI Receptionist for Law Firms | Live Demo | Softcery](https://softcery.com/demos/legal-receptionist): AI intake receptionist for law firms: answers intake calls, qualifies cases, routes callers, and follows the firm's process. Recorded demo included.
- [Conversational AI Hardware | On-Premise Server | Softcery](https://softcery.com/hardware): Reference configs that run the conversational AI stack on hardware in your building. No cloud dependency, no per-minute fees, no data leaving the site.
- [Conversational AI Knowledge Base | Softcery Lab](https://softcery.com/lab): Field notes on conversational AI: architecture, cost, and shipping voice and text agents to production. What we learned building agents that run for real customers.
- [Services | Conversational AI Consulting | Softcery](https://softcery.com/services): Advise, Deploy, Build, Operate: consulting, production deployment, custom engineering, and operations for conversational AI on B2B software platforms.
- [Conversational AI Stack | Self-Hosted, Licensed Source | Softcery](https://softcery.com/stack): The conversational AI stack under license: agent runtime, speech, open-weight models, transport, connectors, reliability. Self-hosted on your hardware, no per-minute fees, full source.

## Cases

- [Amazon Product Optimization: AI Analyst | Softcery](https://softcery.com/cases/amazon-product-optimization): How Softcery built an AI system for Vive Health that optimizes hundreds of Amazon products, diagnosing why performance changed and recommending the next action.
- [Vive Health: AI Agent Resolves Cases End-to-End | Softcery](https://softcery.com/cases/vive-agent): How Softcery built an AI agent for Vive Health that turns every request into a structured case, drafts the next action, and automates with human approval.
- [Bullseye: B2B Visitor Identification | Softcery](https://softcery.com/cases/bullseye): How Softcery built Bullseye from scratch, pivoted it to visitor identification, and rebuilt it after acquisition – two years, two owners, one team.
- [AI Content Marketing Platform Case Study | Softcery](https://softcery.com/cases/ai-content-marketing-platform): How Softcery designed an end-to-end AI platform that generates Instagram content, schedules posts, and correlates engagement with actual sales.
- [B2B Lead Generation Platform Case Study | Softcery](https://softcery.com/cases/lead-generation-platform): How Softcery engineered a lead generation SaaS with map-based prospecting, multi-channel campaigns, and a protected 5M+ record UK business database.
- [AI Marketing Consultant Case Study | Softcery](https://softcery.com/cases/ai-marketing-knowledge-assistant): How Softcery helped a marketing agency productize its expertise: an AI consultant built from 100+ articles delivers their methodology 24/7.
- [AI Build Pack Parser: PDF to Data | Softcery](https://softcery.com/cases/ai-based-build-pack-parser): How Softcery built an AI vision system that extracts network topology from complex PDF build packs, turning hours of manual transcription into automated JSON.
- [AI Lookalike Company Search Case Study | Softcery](https://softcery.com/cases/ai-similarity-search): How Softcery built similarity search using multi-vector AI to find records by meaning – not just keywords.
- [CRM AI Agent: Context-Aware Sales Chat | Softcery](https://softcery.com/cases/crm-ai-agent): How Softcery built an AI chat interface that understands where questions come from – delivering instant answers without users explaining context.
- [Financial Advisory AI: SOA Document Generation Demo | Softcery](https://softcery.com/cases/financial-advisory-ai-document-generation): A Softcery demo build: AI document generation that drafts a Statement of Advice from intake data in under 2 minutes.
- [STRAI: AI Guest Messaging for Airbnb Hosts | Softcery](https://softcery.com/cases/str-ai): How Softcery built an AI messaging agent that answers Airbnb guest questions 24/7, with per-listing knowledge, configurable personality, and human handoff.
- [Casegen Call Evaluation: AI Call Quality | Softcery](https://softcery.com/cases/casegen-call-evaluation): How Softcery built Casegen's post-call system: instant assessment of every AI voice conversation, with automated quality scoring and lead qualification.
- [Casegen Outbound Calls: AI for Law Firms | Softcery](https://softcery.com/cases/casegen-outbound-calls): How Softcery built Casegen's outbound calling system for persistent client follow-up, medical provider coordination, and relationship maintenance at scale.
- [Casegen AI: Voice Agents for Law Firm Intake | Softcery](https://softcery.com/cases/casegen-ai): How Softcery built Casegen's AI voice agents that handle 24/7 legal intake - with attorney-level questioning, multilingual support, and zero missed leads.
- [Proximo AI: Prototype to Production Coach | Softcery](https://softcery.com/cases/proximo-ai): How Softcery turned a broken prototype into a working AI career coaching platform in 4 weeks – and built the foundation for a year of growth.
- [AI Meeting Agent: Voice Bot Joins Calls | Softcery](https://softcery.com/cases/meeting-agent-ai): How Softcery built a real-time voice AI agent that joins video meetings, processes conversation with GPT-4o, and responds live via Pipecat and Attendee API.
- [Vive Health: AI Customer Support Case Study | Softcery](https://softcery.com/cases/vive-health): How Softcery built an AI support agent for Vive Health that handles thousands of monthly tickets, with Odoo integration, policy compliance, and quality gains.
- [UpSkill Q&A AI Assistant: Compliance AI Case Study | Softcery](https://softcery.com/cases/upskill-ai): How Softcery helped build the UpSkill Q&A AI Assistant, an AML/CFT answering tool where every answer has to trace back to a source.

## Voice AI Knowledge Base

- [Hosting Nemotron ASR Streaming 0.6B for a Voice Agent on a CPU](https://softcery.com/lab/hosting-nemotron-asr-streaming-on-a-cpu): Serving nvidia/nemotron-speech-streaming-en-0.6b on a 6-core desktop CPU for a voice agent: a 66 ms final transcript at 12 callers, the chunk size trade, the runtime spin option, 3.1 % word error rate and the break between 24 and 48.
- [Running PhoneLLM Alpha 1 on RTX 3090: VRAM, Latency, Callers](https://softcery.com/lab/running-phonellm-alpha-1-on-rtx-3090): Serving PhoneLLM Alpha 1 for a voice agent on a rented RTX 3090: the quant that fits, a 47.6 ms prompt-eval floor, checkpoints on a Mamba2 hybrid, barge-in reach, the cold-join stall and 24 callers per card.
- [Running Gemma 4 26B-A4B on RTX 3090: Setup, Latency, Benchmarks](https://softcery.com/lab/running-gemma-4-26b-a4b-on-rtx-3090): Self-hosting Gemma 4 26B-A4B for a live voice agent on a rented RTX 3090: VRAM math, the chat template that breaks the cache, a silence bug, barge-in limits, and the turn budget.
- [Softcery Stack Reference: Reliability Layer](https://softcery.com/lab/voice-agent-observability): Reference documentation for the reliability layer of the Softcery conversational AI stack: trace schema, latency model, failure taxonomy, alerting, health probes, storage, known limitations.
- [Fine-Tuning STT for a Voice Agent](https://softcery.com/lab/fine-tune-stt-voice-agent-domain-accuracy): How to fine-tune STT for voice agents: exhaust keyword boosting and prompting, pick a streaming base model (Nemotron, Qwen3-ASR), build training audio from real calls, and gate the release on entity accuracy.
- [Fine-Tuning the LLM in a Voice Agent](https://softcery.com/lab/fine-tune-llm-voice-agent-accuracy-tool-use): Fine-tuning the LLM in a voice agent: build training data from real calls, start with no archive, train tool calling on small open models, and gate the release.
- [Which Layer of a Voice Agent to Fine-Tune](https://softcery.com/lab/how-to-fine-tune-voice-agents): Fine-tuning a voice agent, layer by layer: which symptoms belong to STT, the LLM, or TTS, what fixes each without training, and when fine-tuning actually pays.
- [AI Voice Agent Cost Per Minute at Scale (2026)](https://softcery.com/lab/ai-voice-agent-cost-per-minute-at-scale): The advertised $0.05/min is one meter of several. Model the real per-connected-minute bill across three stacks and three volume tiers with current 2026 rates.
- [What It Costs to Build an AI Voice Agent (2026)](https://softcery.com/lab/how-much-does-it-cost-to-build-an-ai-voice-agent): Voice agent quotes run from $5K to $500K for what sounds like the same product. Here is the itemized bill of materials, priced line by line at 2026 rates.
- [Self-Hosted Voice AI: Cost & GPU Math (2026)](https://softcery.com/lab/self-hosted-voice-ai-stack): A full-stack guide to self-hosting voice AI: four deployment tiers, GPU sizing per concurrent call, the real quality gap, and when a BAA is cheaper.
- [Self-Hosted Voice AI: When It's Worth It (2026)](https://softcery.com/lab/self-hosted-voice-ai-stack-brief): The short version: four deployment tiers, GPU sizing per concurrent call, the real quality gap, and when a BAA beats owning GPUs.
- [Self-Hosted Voice AI: Cost & Feasibility (2026)](https://softcery.com/lab/self-hosted-voice-ai-stack-essence): Self-hosting voice AI: four deployment tiers, GPU sizing per concurrent call, the real quality gap, and when a BAA is cheaper than owning GPUs.
- [Pay-by-Bank & Agentic Commerce: ACP, AP2, MCP, UCP](https://softcery.com/lab/agentic-commerce-protocols-pay-by-bank-checkout): For Pay-by-Bank providers building agentic checkout: what blocks native ChatGPT, Gemini, and Claude distribution, what the protocols solve, what ships now.
- [Agentic Commerce: Selling Through ChatGPT & Gemini](https://softcery.com/lab/how-ai-commerce-payments-work): How agentic commerce works after OpenAI's 2026 checkout pivot: what's live in ChatGPT, Gemini, and Claude, and how payment and liability flow.
- [How to Make Your Store Visible to AI Shopping Agents](https://softcery.com/lab/how-to-make-your-ecommerce-store-visible-in-ai-shopping): Making your store visible to AI shopping agents: where you can connect today, what's gated, and how to prep catalog data, feeds, structured data, and checkout.
- [Self-Hosting Streaming STT for Voice Agents](https://softcery.com/lab/self-hosting-streaming-stt-for-voice-agents): Measured: 83 ms end-of-speech latency, 6.9–8.4% WER, and 30x lower cost than Deepgram at GPU saturation. What self-hosting streaming STT actually takes.
- [Self-Hosting TTS for Voice Agents: Open Models](https://softcery.com/lab/self-hosting-tts-for-voice-agents): Which open TTS model can a voice agent actually ship? Streaming, batching, cloning, and license filters applied, with measured cost vs ElevenLabs and Cartesia.
- [EU Voice AI Regulations 2026: AI Act, GDPR & Call Recording](https://softcery.com/lab/eu-voice-ai-regulations-founders-guide): EU voice AI regulations 2026: AI Act Article 50 disclosure from 2 Aug 2026, GDPR voiceprints, ePrivacy robocall opt-in, and call-recording consent by country.
- [Middle East Voice AI Regulations 2026: UAE, Saudi, GCC](https://softcery.com/lab/middle-east-voice-ai-regulations-founders-guide): Voice AI regulations in the Middle East 2026: UAE, Saudi Arabia, Israel, and the GCC. Criminal call-recording rules, data localisation, voiceprint permits.
- [UK, Switzerland & Non-EU Europe Voice AI Rules 2026](https://softcery.com/lab/non-eu-europe-voice-ai-regulations-founders-guide): Voice AI regulations in non-EU Europe 2026: UK, Switzerland, Turkey, Ukraine – call-recording consent, EU adequacy, and the EU AI Act's extraterritorial reach.
- [Multilingual & Code-Switching Voice AI: Engineering Guide](https://softcery.com/lab/multilingual-code-switching-voice-agents): The engineering guide to multilingual voice AI and code-switching agents: ASR architecture, speech-native models, cross-lingual voice cloning, and prosody.
- [Voice Agent Latency Budget: Mic to Speaker](https://softcery.com/lab/voice-agent-latency-budget-microphone-to-speaker): The full engineering budget for voice agent latency: every component in milliseconds from microphone to speaker, with the techniques that hit sub-800 ms.
- [Voice Prompt Engineering for AI Agents](https://softcery.com/lab/voice-agent-prompt-engineering): The prompt engineering playbook for streaming voice agents: voice-first formatting, number reliability, persona, and the never-claim-an-action-done rule.
- [Voice AI Telephony Stack: SIP, WebRTC, PSTN](https://softcery.com/lab/voice-agent-telephony-stack-sip-webrtc-sbc-codec): The telephony layer beneath production voice AI: SIP and PSTN integration, carrier comparison, codec choice, STIR/SHAKEN, and the IVR-replacement path.
- [AI Voice Agents for Personal Injury Intake](https://softcery.com/lab/ai-voice-agents-for-personal-injury-intake): AI voice agents for personal injury intake: architecture, bilingual requirements, compliance, and a Softcery case study on the missed-call problem.
- [AI Call Center Automation: Actionable Playbook for 2026](https://softcery.com/lab/ai-call-center-automation-playbook-2025): Deploying AI voice agents in real call centers: use cases, performance metrics, compliance, and tech-stack choices, built for scale and real impact in 2026.
- [AI Voice Agents for Travel: Architecture & GDS Guide](https://softcery.com/lab/ai-voice-agents-for-travel-agencies-selection-integration-guide): AI voice agents for travel agencies: real-time vs STT-LLM-TTS architecture, GDS/OTA integration, PCI and GDPR compliance, and the HotelPlanner case study.
- [Custom AI Voice Agents: The Ultimate Guide (2026)](https://softcery.com/lab/custom-ai-voice-agents-the-ultimate-guide): Building custom AI voice agents in 2026: cascaded STT-LLM-TTS vs speech-to-speech, platforms, tooling, and the build-vs-buy threshold (~10K min/month).
- [Best LLMs for Voice Agents 2026: 11 Compared](https://softcery.com/lab/ai-voice-agents-choosing-the-right-llm): Choosing an LLM for voice agents in 2026: 11 models compared on latency, accuracy, and cost, why reasoning modes are voice-unviable, and picks by use case.
- [Real-Time vs Turn-Based Voice Agents 2026](https://softcery.com/lab/ai-voice-agents-real-time-vs-turn-based-tts-stt-architecture): Voice agent architectures in 2026: chained STT-LLM-TTS, half-cascade speech-to-speech, and native audio models, with cost, latency, and telephony trade-offs.
- [10 AI Voice Agent Development Companies Compared](https://softcery.com/lab/top-10-ai-voice-agent-development-companies): 10 AI voice agent development companies compared: custom devs, consultancies, and platforms, with evaluation criteria and a platform-vs-custom framework.
- [9 AI Agent Observability Platforms Compared 2026](https://softcery.com/lab/top-8-observability-platforms-for-ai-agents-in-2025): 9 AI agent observability platforms compared for 2026: Phoenix, LangSmith, Langfuse, Logfire, and more, across deployment, integration, pricing, and use case.
- [14 AI Agent Frameworks Compared (2026)](https://softcery.com/lab/top-14-ai-agent-frameworks-of-2025-a-founders-guide-to-building-smarter-systems): 14 AI agent frameworks compared for 2026: LangChain, LangGraph, CrewAI, OpenAI SDK, and more, with pros/cons, benchmarks, and picks by use case and team size.
- [Why AI Agents Fail in Production: 6 Patterns & Fixes](https://softcery.com/lab/why-ai-agent-prototypes-fail-in-production-and-how-to-fix-it): Six architecture patterns that break AI agents in production: overloaded prompts, PoC architecture, brittle tools, no tests, no observability, all-in rollouts.
- [Why Voice Agents Sound Great in Demos but Fail in Production](https://softcery.com/lab/why-voice-agents-sound-great-in-demos-but-fail-in-production): Why AI voice agents pass demos but fail in production: the technical and business gaps companies hit, and how to make your agent deliver in real calls.
- [Deploying & Scaling Voice Agents: POC to Production](https://softcery.com/lab/deployment-scaling-voice-agents-which-capabilities-when): Deploying and scaling AI voice agents across 4 phases (POC, Pilot, MVP, Full): platform vs custom, capability matrix, SLOs, cost controls, build-vs-buy.
- [Agentic Coding: Claude Code & Cursor Guide](https://softcery.com/lab/softcerys-guide-agentic-coding-best-practices): Agentic coding with Claude Code and Cursor: context files, working memory, Skills and Slash Commands, Subagents, Hooks, MCP, and the current tool landscape.
- [12 Voice Agent Platforms Compared (2026)](https://softcery.com/lab/choosing-the-right-voice-agent-platform-in-2026): 12 voice agent platforms compared for 2026: Vapi, Ultravox, Retell, Bland, LiveKit, ElevenLabs, and more, with pricing, architecture, and a decision framework.
- [SOC 2 for Voice AI Agents: 8 Steps to Compliance](https://softcery.com/lab/soc-2-essentials-for-voice-ai-agents): SOC 2 for AI voice agents: the five trust principles, why enterprises require it, and 8 implementation steps that turn security into a sales asset.
- [US Voice AI Regulations 2026: TCPA, BIPA, HIPAA](https://softcery.com/lab/us-voice-ai-regulations-founders-guide): US voice AI regulations for 2026: the federal floor (FTC, COPPA, TCPA, TAKE IT DOWN) and the state mosaic (BIPA, Colorado, Texas), plus a 5-step plan.
- [Testing Voice Agents: Methods, Metrics, Tools](https://softcery.com/lab/ai-voice-agents-quality-assurance-metrics-testing-tools): Testing AI voice agents for production: methods (functional, UX, performance, accuracy), metrics (FCR, WER, latency, CSAT), monitoring, and the tooling.
- [STT and TTS for Voice Agents: 14 Providers Compared](https://softcery.com/lab/how-to-choose-stt-tts-for-ai-voice-agents-in-2025-a-comprehensive-guide): STT and TTS for voice agents compared: 14 providers including ElevenLabs, Deepgram, AssemblyAI, and Cartesia, with WER accuracy, latency, pricing, and criteria.
- [Voice Agent Architecture | Softcery Lab](https://softcery.com/lab/architecture): The Architecture drawer of the Softcery Lab corpus: real-time vs turn-based pipelines, telephony stack, latency budget, and self-hosting the voice stack.
- [Agentic Commerce | Softcery Lab](https://softcery.com/lab/commerce): The Commerce drawer of the Softcery Lab corpus: agentic commerce, AI payments, and getting stores seen by AI shoppers.
- [Voice AI Compliance | Softcery Lab](https://softcery.com/lab/compliance): The Compliance drawer of the Softcery Lab corpus: voice AI regulations across the US, EU, UK, and Middle East, plus SOC 2.
- [Voice Agent Economics | Softcery Lab](https://softcery.com/lab/economics): The Economics drawer of the Softcery Lab corpus: per-minute cost, build cost, and when self-hosting the voice stack pays.
- [Legal AI | Softcery Lab](https://softcery.com/lab/legal-ai): The Legal AI drawer of the Softcery Lab corpus: voice and language agents for law firms and legal work.
- [Voice Agent Practice | Softcery Lab](https://softcery.com/lab/practice): The Practice drawer of the Softcery Lab corpus: testing, prompting, deployment, and the failure modes that sink voice agents.