You used to wire the providers, glue the scripts, parse the logs, and fix it yourself. Here AI is the architecture, not a feature — the vault, the connectors, the scheduling, the audit all sit behind it. Competitors bolt AI onto an older core. You can't bolt on a center. The deepest native AI integration in the industry — 10 LLM providers, a complete RAG pipeline, image generation, speech-to-text, text-to-speech, fine-tuning, PII redaction, and output validation. Built into the engine. Not bolted on.
9 Providers | RAG Pipeline | AI Media | PII Redaction | Autonomous Agent | Multi-Channel Outbound Notifications (8 channels) | 5,000+ Skills | 367-Tool MCP Server
Five Claude tools, each built for a specific AI workflow pattern. From a single prompt to a fully autonomous agent that operates the InTouch AI platform on its own.
Single-turn Claude prompts with extended thinking, vision/image analysis, temperature control, and system prompts. Process text, analyze images, and generate structured output within your automation workflows.
Models: Claude Opus 4, Sonnet 4, Haiku 4.5
Fully autonomous AI agent that operates InTouch AI with 19 built-in tools — list workflows, run workflows, wait for completion, read execution logs, check server status, and more. The agent reasons about what to do and takes action independently.
Agentic tool-use loop with custom tool definitions. Define tools as HTTP endpoints — Claude reasons about which tools to call, InTouch AI auto-executes them, and feeds results back. Up to 10+ iterations per execution.
Bulk processing with the Anthropic Batch API at 50% reduced cost. Process thousands of prompts in a single tool run — ideal for document classification, data enrichment, and content generation at scale.
Multimodal document and image analysis with citation extraction. Analyze PDFs, images, and complex documents. Get structured responses with page references and source citations.
One general InTouch AI does what a rack of specialized older connectors can't — and. Every provider is a first-class tool — scheduled, triggered, monitored, retried, error-handled. Exactly like SQL or FTP. No SDK zoo, no adapter sprawl. Five providers share a common engine with PII redaction and JSON schema validation built in. Swap the provider by changing one field; the workflow stays the same.
GPT-4o with JSON response format, Azure compatibility, and organization support. The reference provider for the OpenAI-compatible engine.
French AI with Codestral for code generation, mistral-large for reasoning, mistral-small for fast classification. Strong multilingual capabilities.
Sub-second inference on custom LPU hardware. Free tier available. Ideal for real-time classification, high-throughput batch processing, and cost-sensitive pre-processing.
Chain-of-thought reasoning with deepseek-reasoner. Dedicated code model. Competitive pricing for general-purpose work.
Less restrictive content policies. Provider fallback when others decline requests in specific domains.
Gemini 2.5 Pro and Flash with thinking mode. Google's AI capabilities directly in your automation.
Free, private, no API key. Any installed model — Llama, Mistral, CodeLlama, Gemma, Phi, Qwen. Zero data leaves your network.
One key reaches every model. The live catalog across vendors, addressed by slug — anthropic/claude-sonnet-4.5, openai/gpt-5.1. Switching models is editing a string.
500,000+ open-source models. 10 NLP tools: classification, NER, summarization, translation, QA, embeddings, and more. Free tier for most models.
Before any prompt leaves your network, sensitive data is replaced with [REDACTED]. Five built-in patterns — SSN, email, phone, credit card, IPv4 — plus custom regex. Works identically across OpenAI, Mistral, Groq, DeepSeek, and xAI. Redaction count logged in the audit trail.
This is the contract's "what to do when it doesn't work" clause, made intelligent. Set a JSON Schema on any AI tool. The engine validates every response — required fields, type checking, structure. Validation fails, it reads the error, re-sends the prompt with the error appended, and corrects itself. Up to 6 attempts. Structured JSON that fixes itself. No human in the loop. A dumb rule could never do that.
A complete RAG pipeline as native tools. No LangChain. No Python scripts. No glue code to maintain at 2am. Chunk documents, generate embeddings, store vectors, retrieve by semantic similarity, generate grounded answers — every step a schedulable, auditable workflow tool under enterprise governance.
Split documents into right-sized pieces for embedding. 6 strategies: tokens (with overlap), sentences, paragraphs, lines, pages, or custom regex. Configurable size, overlap, and metadata. Supports txt, markdown, HTML, CSV, JSON.
Convert text to dense vectors via OpenAI-compatible API. Texts with similar meaning produce similar vectors — "cancel my subscription" matches "stop my monthly plan." Alternative: Hugging Face feature-extraction for free embeddings.
Upsert, query, delete across Pinecone, Weaviate, Chroma, Qdrant, pgvector, Milvus, or a zero-dependency local file store with cosine similarity. Metadata filtering, namespace isolation, top-K retrieval.
AI Assistant grounded in your documents — load handbooks, runbooks, policies into the vector store. The assistant cites your actual content in delivered notifications instead of hallucinating. Workflow failure diagnosis — embed error logs, query for similar past failures, auto-notify with context. Intelligent file routing — files land in a folder, get classified semantically, route to the right workflow. Compliance audit search — "credential changes in Q1" finds entries logged as "rotated token" or "updated API key." The local vector store means Personal edition runs the entire pipeline with nothing beyond Java.
Four more AI capabilities, all on one OpenAI credential. One connection, five capabilities. No new wiring.
DALL-E 3, DALL-E 2, and gpt-image-1. Create images from text prompts for reports, marketing, notifications, and content pipelines. HD quality, multiple sizes, automatic file download.
Transcribe audio files (mp3, wav, m4a, webm) to text, SRT subtitles, or VTT. Three models: Whisper-1, GPT-4o-transcribe, GPT-4o-mini-transcribe. Auto language detection. Vocabulary guidance for domain terms.
9 voices (alloy, ash, coral, echo, fable, onyx, nova, sage, shimmer), 6 audio formats, 0.25x-4x speed control. Generate audio notifications, accessibility content, podcast material, IVR messages.
Create, monitor, cancel, and list OpenAI fine-tuning jobs and models. Train custom models on your data for domain-specific classification, consistent formatting, and structured extraction. Department & Enterprise editions.
Pure AI on one end, fully deterministic on the other — you set the dial per use case. Start a workflow as freeform AI judgment, then graduate it to a hard when: rule once it earns trust: zero AI cost, identical every time, fully audited. Nobody else gives you the dial. Monitors and the if task are where you turn it.
A Monitor runs a check: tool on a schedule and evaluates explicit when: arms against its output. The first matching arm fires one of five actions — notify, run_tool, run_skill, run_jobfile, nothing. Author in IML, install from the hub, or describe in your own language. Deterministic by default; an optional ai: arm lets the AI assistant evaluate freeform conditions when the rule is genuinely fuzzy. Full Monitor docs →
if Task — Workflow-Local Flow Control NEWThe same arm grammar Monitors use, scoped to a single workflow's task list, plus two extra flow-control verbs: goto (jump forward by sequence) and skip_remaining (halt the rest of the workflow). Same five action verbs Monitors have (notify, run_tool, run_skill, run_jobfile, nothing) plus those two = seven total. Captures intermediate outputs with as: <name> so later actions in the same arm reference {{name.field}}. Replaces the older per-task runif field. If-task docs →
Create alert objects with subject, body, and HTML/text formatting. Link alerts to schedules, trigger files, and Monitors. When any linked trigger fires, the alert is sent to every configured channel simultaneously — up to 8 channels (5 on Personal/Team: Email, Slack, Discord, Telegram, SMS — plus WhatsApp, Microsoft Teams, LINE on Department/Enterprise).
Compiled tools, AI-powered skills, and declarative workflow definitions in one system. No other automation platform does this. InTouch AI does.
Tools are the building blocks. Two types, pick the one the workflow needs:
Full JVM power with custom connections, complex operations, and dependencies. SQL, FTP, Essbase, TM1, AWS, Google Workspace, Docker, MongoDB, Git, LDAP, PDF, Excel, and more. Dynamic schema-driven UI forms — the tool defines its schema, the UI renders it automatically.
Zero-compile tool creation. Define steps declaratively — HTTP calls, AI reasoning, file operations, data transforms, conditions, loops. Perfect for rapid prototyping and lightweight integrations. Same execution engine, same UI, same scheduling.
Skills tell the AI what to accomplish with InTouch AI tools. Write a SKILL.md file, declare the tools it needs, and the AI orchestrates the rest. You describe intent. InTouch AI does the work.
Native markdown skills with YAML frontmatter. Declare required tools and inputs. The AI reads the instructions, validates tool availability, and orchestrates execution. Chainable — skills can invoke other skills.
Use @ followed by the skill name for auto-resolution, @intouch:name for InTouch AI-native skills, or @claw:name for OpenClaw skills. No collisions. Full control over which provider runs.
Plus access to ClawHub (5,000+ OpenClaw skills) and the InTouch Hub. Auto-installs on first use. Invoke from the built-in AI assistant UI, the InTouch PWA, the Android app, the REST API, or schedule as a workflow tool. Two skill ecosystems, one platform.
Whole multi-step workflows in IML. Version-control your automation in git. Deploy across environments. Not one UI click required.
Name, description, tools with credentials, operations, and properties. Tool dependencies with after:, failure handling (onFailure: stop|continue), and retry logic — all in a single IML file.
Reference published properties from prior tools: {{extract.rowCount}}, {{transform.outputFile}}. Data flows between tools automatically. Chain SQL exports to S3 uploads to email notifications.
Validate, install, run, delete, and reload Workflow Files via REST API. Server validates tool availability, credential existence, and dependency graph before accepting. Same execution engine as UI-built workflows.
Alerts, scheduled reports, and workflow results land with your contacts across eight outbound channels. Every channel pulls its credentials by name from the encrypted vault and audit-logs every send. Nothing leaks, nothing goes untracked.
SMTP delivery. Notifications, scheduled reports, alert digests, attachments.
Bot token posts to channels and DMs. Block Kit formatting. Per-channel thread support.
Bot posts to channels and DMs via Gateway. Embeds, mentions, threaded replies.
Bot API sendMessage. Outbound HTTPS only. Group, channel, or one-to-one delivery.
Twilio, Telnyx, and Plivo outbound SMS for phone-number recipients.
Meta Cloud API for outbound WhatsApp messages to verified numbers.
Bot Framework outbound posts to Teams channels via webhook.
LINE Messaging API outbound push to LINE Official Accounts.
Every InTouch AI user gets access to a built-in AI assistant directly in the UI (Ctrl+Enter), in the InTouch PWA, in the InTouch Android app, or through the REST API. Conversational authoring is the primary UI — describe what you need and the assistant builds it. More than a chatbot: it can create workflows, run tools, manage credentials, invoke skills, and set up schedules using tool_use. Configurable provider (Anthropic, OpenAI, Google Gemini, Mistral, DeepSeek, xAI, Hugging Face, Ollama — 8 in total), model, temperature, and system prompt — set once at the server level, available to all authenticated users. All actions respect the user's access control permissions. Swap to Ollama locally for zero-cost, private inference with no data leaving the network — the AI assistant runs very well on a local open model like qwen2.5.
The InTouch MCP (Model Context Protocol) server connects Claude Code directly to InTouch AI, exposing hundreds of tools across the complete REST API:
40+ tools for CRUD, schedules, triggers, execution
Tools for database, FTP, SSH, and cloud connections
Tools for execution logs and history
20+ tools for status, control, and sessions
Tools for access control role management
Tools for server settings and tool management
Tools for schedules, file triggers, and Monitors
Tools for backup, restore, and migration
With the MCP server, you can tell Claude Code: "Create a workflow that exports the sales table every morning at 6am and emails the CSV to the finance team." Claude uses the hundreds of tools to configure everything — credentials, tools, schedules, contacts — without you touching the UI.
So the same server has two AIs pointed at it: an external agent that builds, and a built-in assistant that runs. We asked both the same shop-owner question and published what each one said — the clearest way to see which half does what.
No other automation platform offers 10 native LLM providers with PII redaction and output validation built in. Swap providers by changing one field. If one provider refuses a domain, switch to another — or to Ollama for zero restrictions.
Complete RAG pipeline as native tools — chunk, embed, store, retrieve, generate. 7 vector store providers including a zero-dependency local option. Scheduled, audited, governed. Not a Python script with a scheduled script.
Chain RAG retrieval with SQL exports, AWS uploads, image generation, and speech synthesis — in a single workflow. Image gen + TTS in one workflow. Embed docs on a schedule. Transcribe audio and analyze with Claude. No other platform does this.
Sensitive data never reaches the AI provider. SSN, email, phone, credit card, IPv4 — all redacted automatically before the prompt leaves your network. Custom regex for industry-specific patterns. Enterprise compliance from day one.
Text generation, image creation, speech-to-text, text-to-speech, document analysis, fine-tuning, and 500K+ Hugging Face models — all as governed workflow tools. Not separate products. One platform, one credential vault, one audit trail.
strongly encrypted credentials with access control. Full audit trail. Enterprise governance on every AI operation. And the Personal edition is completely free — no time limits, no feature gates on AI.
10 LLM providers. Complete RAG pipeline. Image generation, speech-to-text, text-to-speech. PII redaction. Output validation that reads its own failures and fixes them. AI at the center — not bolted onto an older core, because you can't bolt on a center. The deepest AI integration in any automation platform.
Contact Blue Isle Compare Editions