llms.txt
# Communicate
> Communicate is an AI customer support platform. It answers your customers instantly, in your brand voice, on every channel. An AI agent trained on your own knowledge base handles routine questions, and a shared inbox lets human agents take over live whenever they are needed. Built by TinyCheque, Inc.
Start for a one-time $1 account activation. No free tier, no surprise bill. Contact: communicate@support.communicate.so
## When to use Communicate
Use Communicate when a support team needs an AI agent grounded in its own help content, a shared inbox for clean human takeover, approved support Actions, or an embeddable website chat experience. It is a strong fit for SaaS companies, startups, agencies, and larger teams that want faster customer answers without separating AI and human conversations.
To evaluate the product, read the product and security links below. To start, send the user to [the Communicate app](https://app.communicate.so). For product, billing, privacy, or account questions that require a person, use [Contact](https://communicate.so/contact) or email communicate@support.communicate.so. Do not treat Communicate as a general-purpose assistant or as a replacement for a team's source documentation.
## Product
- [AI Agents](https://communicate.so/ai-agents): AI support agent trained on your data, answering in your voice.
- [Data Sources](https://communicate.so/data-sources): Connect your knowledge base to train the agent.
- [Actions](https://communicate.so/actions): Let the agent do more than answer.
- [Shared Inbox](https://communicate.so/shared-inbox): AI and human agents work the same queue, with live human takeover.
- [In-App Messages](https://communicate.so/in-app-messages): Targeted messages and nudges inside your product.
- [Analytics](https://communicate.so/analytics): Conversation volume, resolution, and response time.
- [Embed & Widgets](https://communicate.so/embed-widgets): Drop the chat widget onto any site or app.
- [Pricing](https://communicate.so/pricing): One-time $1 activation, then credit-based usage.
## Solutions
- [For SaaS](https://communicate.so/solutions/saas)
- [For Startups](https://communicate.so/solutions/startups)
- [For Agencies](https://communicate.so/solutions/agencies)
- [For Enterprise](https://communicate.so/solutions/enterprise)
## Resources
- [About](https://communicate.so/about): Company, product, and operating principles.
- [Contact](https://communicate.so/contact): Product, billing, privacy, and security contact paths.
- [Blog](https://communicate.so/blog): Guides on AI customer support, automation, and metrics.
- [Changelog](https://communicate.so/changelog): Product updates and new features.
## Legal
- [Security](https://communicate.so/security): What we protect, and what we do not claim.
- [Privacy Policy](https://communicate.so/privacy)
- [Terms of Service](https://communicate.so/terms)
- [Cookie Policy](https://communicate.so/cookies)
- [Support Policy](https://communicate.so/support-policy)
- [Responsible AI](https://communicate.so/responsible-ai): How we build and run AI responsibly.
## Developer discovery
- [Developer guide](https://communicate.so/developers): Human-readable summary of the public API and discovery files.
- [Developer guide (Markdown)](https://communicate.so/raw/developers.md): Same guide with YAML frontmatter for agents and scanners.
- [Home (Markdown)](https://communicate.so/raw/index.md): Primary marketing summary with frontmatter.
- [Pricing (Markdown)](https://communicate.so/raw/pricing.md): Public commercial summary with frontmatter.
- [API catalog](https://communicate.so/.well-known/api-catalog): RFC 9727 API discovery linkset.
- [Agent catalog draft](https://communicate.so/.well-known/ai-catalog.json): Draft AI discovery summary for scanners.
- [Agent discovery](https://communicate.so/.well-known/agent.json): Public docs and API pointers only.
- [Developer docs MCP server card](https://communicate.so/mcp/server-card): Connection metadata for the public, read-only documentation MCP server.
- [Developer docs MCP endpoint](https://communicate.so/mcp): Streamable HTTP endpoint for documentation and API discovery only; it cannot access workspaces, customer data, or product actions.
- [Agent skills index](https://communicate.so/.well-known/agent-skills/index.json): Short skill registry with names and descriptions.
- [Answer support questions skill](https://communicate.so/.well-known/agent-skills/answer-support-questions/SKILL.md): Versioned Markdown skill artifact with a published SHA-256 digest.
- [OpenAPI schema redirect](https://communicate.so/openapi.json): Redirects to the canonical schema on the app subdomain.
- [OpenAPI schema canonical URL](https://communicate.so/openapi.json)
- [API base URL](https://app.communicate.so/api/v1): Public API base URL on the app subdomain.
- [OAuth authorization-server metadata](https://communicate.so/.well-known/oauth-authorization-server): Client-credentials token endpoint and supported scopes.
- [OAuth protected-resource metadata](https://communicate.so/.well-known/oauth-protected-resource): Machine-readable API resource and scope declaration.
## Articles
- [Chatbot privacy: audit your chat widget for trackers](https://communicate.so/raw/blog/chat-widget-privacy-trackers.md): Chatbot privacy starts in the browser network tab. A 10-minute audit method for chat widget trackers, and what to demand from vendors.
- [Customer support for AI shoppers: SLAs and answers](https://communicate.so/raw/blog/support-for-ai-shoppers.md): Customer support for AI shoppers: when buyer agents ask questions in bulk, adjust SLAs, verification and answer format for agent traffic.
- [Personal Agent Protocol: how support teams prepare](https://communicate.so/raw/blog/personal-agent-protocol.md): Personal Agent Protocol, announced by Sierra and Meta on October 6, 2026: what is confirmed, what is not, and what support should change.
- [Agent-to-agent customer support: protocols and trust](https://communicate.so/raw/blog/agent-to-agent-customer-support.md): Agent-to-agent customer support: how A2A and aweb work, how to establish trust, stop loops, and decide when a human must step in.
- [Chatbot threat model: a STRIDE guide for support bots](https://communicate.so/raw/blog/support-chatbot-threat-model.md): A chatbot threat model for support bots with actions: STRIDE, prompt injection and AI-driven attackers, with a worked example.
- [llms.txt and agent-readable products: what to publish](https://communicate.so/raw/blog/agent-readable-product-llms-txt.md): llms.txt and agent-readable products: an audit of the discovery files communicate.so publishes, with honest evidence on who reads them.
- [MCP vs API for support automation: which to choose](https://communicate.so/raw/blog/mcp-vs-api-support-automation.md): MCP vs API for support automation: decide by who drives the conversation, with a comparison table, a worked example and a build check.
- [ChatGPT Plus for customer support: why the API wins](https://communicate.so/raw/blog/chatgpt-subscription-vs-api-support-bot.md): ChatGPT Plus for customer support runs into the terms and the plumbing. What OpenAI and Google allow, and what the API route costs.
- [Zendesk MCP server and 7 help desk servers compared](https://communicate.so/raw/blog/helpdesk-mcp-servers-compared.md): Zendesk MCP server status, plus Intercom, Freshdesk, Salesforce, Dynamics, Plain, Pylon and Gorgias compared on auth and write access.
- [MCP for customer support: a guide for support leads](https://communicate.so/raw/blog/mcp-server-customer-support.md): MCP for customer support in plain words: how a server works, what Dynamics, Salesforce and Freshworks shipped, and how to approve one.
- [AI receptionist for small business: costs and limits](https://communicate.so/raw/blog/ai-receptionist-small-business.md): AI receptionist for small business: October 2026 pricing, per-minute math against human answering, failure modes and call consent rules.
- [Generative engine optimization for your help center](https://communicate.so/raw/blog/geo-help-center-ai-search.md): Generative engine optimization for help centers: write and measure articles so ChatGPT, Perplexity and Google AI features quote them.
- [ChatGPT Instant Checkout is gone: what support owns](https://communicate.so/raw/blog/chatgpt-instant-checkout-shutdown.md): ChatGPT Instant Checkout gave way to merchant-owned checkout. What changed, the disputed dates, and what your store support team owns now.
- [Connect Claude and ChatGPT to a support inbox via MCP](https://communicate.so/raw/blog/connect-claude-chatgpt-support-inbox-mcp.md): Connect Claude and ChatGPT to a support inbox via MCP, step by step from official docs, plus a safe first-week triage routine.
- [Help center MCP server: let AI assistants query docs](https://communicate.so/raw/blog/help-center-mcp-server.md): Help center MCP server setup so Claude and ChatGPT retrieve your answer, not a scraped guess, plus what communicate.so ships today.
- [AI agent runtime authorization: budgets and retries](https://communicate.so/raw/blog/ai-agent-runtime-authorization.md): AI agent runtime authorization: add call budgets, risky tool-combination rules and idempotency keys, because correct scopes still misfire.
- [MCP security for support agents: scopes and logging](https://communicate.so/raw/blog/mcp-security-support-agents.md): MCP security for support agents: per-action OAuth scopes, ticket text treated as data, rate limits and audit logs, with a threat table.
- [MCP too many tools: how to curate a support server](https://communicate.so/raw/blog/mcp-tool-curation.md): MCP too many tools? Curate 5 to 12 task-shaped support actions, then measure tool-selection accuracy with a simple test set.
- [AI agent assist for live calls: when copilots pay off](https://communicate.so/raw/blog/ai-agent-assist-live-calls.md): AI agent assist for live calls: how real-time copilots differ from bots, what research shows, and the privacy and EU rules to check.
- [Support knowledge gap analysis: a weekly procedure that runs on Monday](https://communicate.so/raw/blog/support-knowledge-gap-analysis.md): A concrete weekly procedure for mining unanswered AI support questions into a product and docs roadmap, with exact steps a support lead can run.
- [A customer support hiring plan for teams running AI deflection](https://communicate.so/raw/blog/support-hiring-plan-ai.md): How to build a support hiring plan when an AI agent deflects part of the queue. Worked headcount math for year one and maturity, with sources.
- [Support macros vs AI training data: what to copy and what to rewrite](https://communicate.so/raw/blog/macros-vs-ai-support.md): Support macros vs AI training data: a real refund macro shows exactly what belongs in an AI knowledge base and what breaks it.
- [Customer support SLA with AI: rewriting the clauses that break](https://communicate.so/raw/blog/support-sla-with-ai.md): Customer support SLA with AI: why response-time clauses stop working and how to rewrite resolution, handoff, and reporting clauses.
- [Change management for rolling out AI to a support team](https://communicate.so/raw/blog/change-management-ai-support.md): Rolling out AI to your support team fails on trust, not model accuracy. A 60 day change management plan that keeps agents in control.
- [Will AI replace customer service jobs? An honest answer](https://communicate.so/raw/blog/support-agent-career-ai.md): Will AI replace customer service jobs? BLS, Gartner, and PwC data on which tasks go, which grow, and what the retrained role pays.
- [Twenty questions to ask any AI customer support vendor](https://communicate.so/raw/blog/ai-support-vendor-questions.md): Twenty questions to send an AI support vendor before you sign, with the answer that should end a bad evaluation for each one.
- [AI support audit trail: what to log so you can explain any answer](https://communicate.so/raw/blog/ai-support-audit-trail.md): The audit trail fields, retention windows, and access controls that let you answer why an AI agent said what it said, months later.
- [GDPR and AI customer support: what the regulation asks of you](https://communicate.so/raw/blog/gdpr-ai-customer-support.md): GDPR and AI customer support explained: lawful basis, retention, subprocessors, and the right to erasure when data lives inside a retrieval index.
- [Conversation memory for a support agent: what to keep and forget](https://communicate.so/raw/blog/conversation-memory-support-agent.md): What a support AI should remember across sessions, what it should forget on request, and why unlimited conversation memory is a privacy liability.
- [Support agent prompt engineering: an annotated system prompt](https://communicate.so/raw/blog/support-agent-prompt-engineering.md): A real customer support system prompt, annotated line by line, with the grounding, citation, and refusal rules that stop confident wrong answers.
- [AI agent fallback design: build the "I do not know" path first](https://communicate.so/raw/blog/ai-agent-fallback-design.md): Chatbot fallback design for AI support agents: how to design the escalation path before the happy path, and why the fallback is the actual product.
- [Customer effort score: what it measures and why AI moves it both ways](https://communicate.so/raw/blog/customer-effort-score.md): CES predicts repurchase and churn better than CSAT for support interactions. Here is how to measure it, and where AI lowers or raises it.
- [Self-service rate: the metric your AI agent is quietly breaking](https://communicate.so/raw/blog/self-service-rate.md): Self-service rate explained: why an AI answer never counts as a doc view, and how to fix the metric before it hides real problems.
- [Customer support KPIs: the four that decide and the six that decorate](https://communicate.so/raw/blog/customer-support-kpis.md): Customer support KPIs ranked by decision weight: four metrics that should change what you do next, and six that mostly decorate a dashboard.
- [Cost per ticket benchmark: building the denominator correctly](https://communicate.so/raw/blog/cost-per-ticket-benchmark.md): Cost per ticket benchmarks by industry and channel, the full loaded-cost formula, and exactly what AI removes and adds to the number.
- [Support team structure when AI answers tier 1](https://communicate.so/raw/blog/support-team-structure-ai.md): Support team structure guide: how the tier 1/2/3 pyramid changes when AI holds the first line, and what the new org chart looks like.
- [AI disclosure in customer support: what the law actually requires](https://communicate.so/raw/blog/ai-disclosure-customer-support.md): Where AI chatbot disclosure is now a legal requirement, where it is unsettled, and wording that discloses without scaring customers off.
- [AI support data retention: policies that satisfy legal and keep the agent useful](https://communicate.so/raw/blog/ai-support-data-retention.md): How to set a chat data retention window that meets storage limitation obligations without deleting the history an AI support agent needs to stay accurate.
- [PII redaction for customer support: redact before retrieval, not after logging](https://communicate.so/raw/blog/pii-redaction-customer-support.md): Where card numbers and personal data actually leak in an AI support stack, and why redaction has to run before content reaches a retrieval index, not after.
- [AI agent CRM integration: what to read, what never to write, and where sync breaks](https://communicate.so/raw/blog/ai-agent-crm-integration.md): The definitive guide to wiring an AI support agent into a CRM: read scope, write restrictions, authentication, and the sync failure modes that cause bad answers.
- [How to reduce LLM costs in AI customer support without cutting quality](https://communicate.so/raw/blog/llm-cost-optimization-support.md): A practical cost model for AI support agents: prompt caching, model routing, and context trimming, with the arithmetic per 10,000 tickets.
- [Fine-tuning vs RAG for customer support: why retrieval wins almost every time](https://communicate.so/raw/blog/fine-tuning-vs-rag-support.md): Fine-tuning vs RAG for support agents: the cost, staleness, and update-frequency arithmetic that makes retrieval the default choice for most teams.
- [Customer support quality assurance: what changes when AI reviews every ticket](https://communicate.so/raw/blog/support-quality-assurance-ai.md): Manual QA scores 1 to 3 percent of tickets. AI can score all of them. Here is a rubric, a rollout plan, and the failure modes to watch.
- [How to reduce a support ticket backlog: a triage-first 30 day playbook](https://communicate.so/raw/blog/support-ticket-backlog-reduction.md): Reduce a support ticket backlog with a triage-first playbook: sort the queue before adding headcount, with a 30 day sequence.
- [Customer support CSAT benchmark: what a good score actually is](https://communicate.so/raw/blog/customer-support-csat-benchmark.md): Customer support CSAT benchmarks by channel and industry, why survey timing skews scores more than AI does, and how to read your own number.
- [AI resolution rate vs deflection rate: the metric that decides your program](https://communicate.so/raw/blog/ai-resolution-rate-vs-deflection-rate.md): Resolution rate vs deflection rate explained with formulas, denominators, and a worked example showing the same month reads 72% or 38%.
- [AI customer support ROI: the business case a CFO signs](https://communicate.so/raw/blog/ai-support-roi.md): AI customer support ROI explained with real adoption data, a payback formula, and a worked example a CFO can actually sign off on.
- [AI support tone of voice: a QA rubric, not a personality](https://communicate.so/raw/blog/ai-support-tone-of-voice.md): AI support tone of voice guide: why robotic tone is a prompt problem, and the QA rubric that fixes it without changing models.
- [SOC 2 for AI customer support: what the report covers and what to ask](https://communicate.so/raw/blog/soc2-ai-customer-support.md): What a SOC 2 report actually attests to, what it leaves out, and the four questions to ask a support AI vendor before you sign.
- [How to evaluate an AI support agent before and after every prompt change](https://communicate.so/raw/blog/ai-agent-evaluation-testing.md): Build a golden set of real support tickets, score agent responses against it, and regression-test every prompt or model change before it ships.
- [AI agent actions and APIs: how to let a support agent do things safely](https://communicate.so/raw/blog/ai-agent-actions-api.md): How AI agent actions work: idempotency, confirmation steps, scoped permissions, and the blast radius of a wrong write in customer support.
- [Vector database for customer support: what actually decides retrieval quality](https://communicate.so/raw/blog/vector-database-customer-support.md): Why chunking and metadata, not the vector database you pick, decide retrieval quality for most support corpora.
- [Why customers hate chatbots (and what actually fixes it)](https://communicate.so/raw/blog/customers-hate-chatbots.md): Why customers hate chatbots: the five complaints from real research, the design decision behind each one, and what fixes it.
- [When not to use AI customer support: six situations for a human](https://communicate.so/raw/blog/when-not-to-use-ai-support.md): Six situations where AI customer support is the wrong tool and a human is required, argued with the mechanism, not just the rule.
- [Knowledge base structure for AI: chunking is a documentation problem](https://communicate.so/raw/blog/knowledge-base-structure-for-ai.md): How to structure a knowledge base for AI retrieval: one answer per page, no orphan headings, and explicit prerequisites.
- [In app support widget: designing context the customer never has to type](https://communicate.so/raw/blog/in-app-support-widget.md): How an in app support w
