Automation you describe in plain English. Tell the AI assistant what you want done; it builds the automation, schedules it, runs it, and shows you proof it ran. Your credentials stay in an encrypted vault the AI cannot read.
The same codebase runs in 24/7 production at Fortune 500 companies. Personal Edition is free forever — no licence key, no trial, no expiry.
Needs Java 17+. On Windows the installer installs Python for you.
Windows — extract install-intouch-ai.zip, double-click install.bat
Linux / macOS
unzip install-intouch-ai.zip -d install-intouch-ai
cd install-intouch-ai
python3 install.py
Log in at http://localhost:2200 (intouch / blueisle), paste an
openrouter.ai key into the Get Started panel, and type what you want.
Put a dollar on the key — a free key is good for a handful of calls and then stops. Or run a local Ollama model instead, with no key and no cost.
Start and stop with python3 bin/start.py and python3 bin/stop.py.
| 35 tool types | SQL (10 relational databases), SSH, FTP/SFTP, rsync, HTTP, AWS (the entire CLI v2), Google Workspace, Docker, the AI family, and more — every one scheduled, retried, logged and alerted the same way. Hundreds more install from the Hub |
| 8 AI providers | OpenRouter (recommended — one key reaches 367 models), Anthropic, OpenAI, Gemini, Mistral, DeepSeek, xAI, and Ollama for fully local |
| 35 credential types | AES-256-GCM vault; secrets are redacted out of every API response, so the AI can use a credential without being given it |
| AI assistant | Reads the real documentation and your real install. Builds workflows, runs tools, explains failures |
| 271-tool MCP server | Drive InTouch from Claude Code or any MCP client. InTouch is an MCP client too |
| Alerts on 4 channels | Email, Slack, Discord, Telegram — outbound (SMS, WhatsApp, Teams and LINE on Department and Enterprise) |
| 7 schedule types | Plus file triggers, folder triggers, AI-condition triggers, and ad-hoc OneShot runs |
| IML | Workflows and tools as code — plain JSON, diffable, reviewable, portable between installs |
| Custom tools | Implement IToolConnector, drop in a JAR, and your tool is a first-class citizen |
| Runs on a Pi | Single JAR, embedded database, zero external dependencies. A Raspberry Pi 5 is enough |
Then you have the authoring half of the problem solved, and this is the other half.
An AI coding agent decides what should happen. It cannot make it keep happening — it cannot wake up at 03:00, wait four hours for a condition without burning context, do the identical thing on the hundredth run, or hold a production credential it has not read. Those are runtime problems, not intelligence problems.
InTouch runs as a service on the same machine as your terminal:
your terminal InTouch AI (background service)
───────────── ────────────────────────────────
Claude Code ──MCP──▶ 271 tools · 489 REST endpoints
authors, schedules · triggers · retries · alerts
proves it, AES-256-GCM credential vault the AI cannot read
hands it over attributed audit log · frozen, zero-cost runs
Point your agent at the MCP server once, then talk:
"Store these API credentials in InTouch, then build a workflow that pulls yesterday's orders into the warehouse." · "Run it once and show me the log." · "Schedule it for 02:00 and alert me on failure." · "Freeze it — no AI in the execution path from here."
Four sentences, and the thing exists, is proven, is scheduled, is monitored, and no longer costs a token to run.
Notes from Claude Code — Anthropic's CLI agent wrote its own account of what it hands over and why, after working inside a running InTouch server.
Most platforms make you choose: hand-write every step, or hand everything to a model. InTouch spans the range, and you pick per workflow.
Enterprises tend to run deterministically in production while using AI from conception through authoring. Newcomers start heavily AI and migrate the parts that matter to determinism.
| Edition | What it adds |
|---|---|
| Personal | Free forever. One user, 25 contacts, everything else unrestricted |
| Solo | Your own named login, unlimited contacts, areas and tracks |
| Team | Up to 5 users, shared workflows, credentials and run history |
| Department | Role-based access control, area isolation, LDAP / Active Directory, CyberArk, import/export |
| Enterprise | Everything, plus TLS and cipher governance, SSO, and live workflow monitoring |
Every edition runs the same code. Personal stays simple by hiding the areas/tracks hierarchy — work lands in a default bucket created at startup. You need a licensed edition when you want named or multiple users, that hierarchy, or governance.
Ask the assistant. It reads the complete documentation and can see what is installed on your machine, so it answers about your install rather than a generic one.
Everything else is at blueisle.com — editions · use cases · architecture · security · comparisons
The platform is commercial; the ecosystem on top of it is open. Build tools against the
MIT-licensed intouch-tool-api, or author workflows and tools as IML, and install them from
anywhere a file lives — a private git repo, an internal artifact server, a coworker's share, or
the Hub. The Hub is a convenience, not a required channel.
Personal Edition is free under the EULA. Solo, Team, Department and Enterprise are commercially licensed — see blueisle.com/editions.html.