| InTouch AI | Zapier |
| What it is | Automation platform with a built-in agentic AI assistant. Runs on your own machine, from a laptop to a Raspberry Pi to a server. | Hosted integration service that connects cloud apps with trigger-and-action chains. |
| Who it is for | One person with a to-do list through to a Fortune 500 IT floor — technical or not. The assistant is the way in. | Non-technical users connecting SaaS apps. Deliberately shallow, deliberately easy. |
| Where it runs | Your hardware, your network. Nothing has to leave the building. Also fine in a cloud VM if you want one. | Their cloud only. There is no self-hosted Zapier. |
| How you pay | Personal Edition free forever, no per-run charge at any volume. Paid editions are licences, not meters. | Per task — every action step that executes is billed. Automation that works more costs more. |
| InTouch AI | Zapier |
| How automation gets built | Describe it in plain English and the assistant builds it. Or write IML directly. Or both — edit what it wrote. | Point-and-click in their web builder. An AI copilot can draft a Zap for you. |
| What the automation IS | IML — plain JSON you own. Diff it, review it in a pull request, commit it, carry it to another install. | A Zap inside their account. Exportable as a template, not as a reviewable artifact. |
| Adding a capability that does not exist yet | Implement one interface, drop in a JAR, and it hot-loads without restarting. It arrives already scheduled, credentialed, retried, logged and alerted. | Build a private integration in their developer platform, or drop into a Code step (JS/Python) inside the sandbox they allow. |
| Configuration screens for your own tools | The Schema Designer renders a tool’s entire configuration form from its schema — including tools you wrote yourself. Nobody writes UI code per tool. | Field definitions in their developer platform render inside Zapier's UI. |
| Reach beyond SaaS APIs | 35 tool types in the free download: 11 SQL databases, SSH, FTP/SFTP, rsync, HTTP, AWS (the entire CLI v2), Google Workspace, OS/spawn, runtime environments, messaging, skills, and the AI family (Anthropic ×5, OpenAI, Gemini, Mistral, DeepSeek, xAI, OpenRouter, Ollama, embeddings, vector store, speech, images). Essbase and TM1 bring it to 37 once licensed. Docker, Git, Excel, PDF, LDAP, MongoDB, Cassandra and thirty more install from the Hub in one click. | Whatever an app connector exposes. Reaching a server, a shell or a database on your own network needs a gateway or a webhook you build. |
| InTouch AI | Zapier |
| Scheduling | 7 native schedule types, plus file triggers, folder triggers and AI-condition triggers. | A schedule trigger on an interval, plus app and webhook triggers. |
| Reacting to a file landing on disk | First-class. A trigger watches a path or folder, with a settle timer so it does not fire on a half-written file. | Cloud storage apps only (Drive, Dropbox…). A local folder needs something you build. |
| Stopping two runs colliding | Collision detection and exclusive locks, so a long run cannot trample the next one. | Not offered. |
| InTouch AI | Zapier |
| Where the AI sits | Inside the engine. The assistant builds tools, workflows, credentials, schedules and monitors, runs them, reads the logs and fixes what broke. | Alongside the product — a copilot that drafts Zaps, AI steps you can add, and agent features. |
| Choosing your own model | 8 providers — OpenRouter (one key, 367 models), Anthropic, OpenAI, Gemini, Mistral, DeepSeek, xAI, and Ollama for fully local. | Their hosted AI, or an OpenAI-style connector you configure. |
| Running with no AI at all | Fully supported and common: AI authors the workflow once, then it runs forever with no model in the execution path and zero AI cost per run. | A Zap without AI steps uses no AI — but the AI never authored a reusable artifact you then own. |
| Cost of AI over time | Zero per run once authored, because the artifact is deterministic. You can also watch spend by provider, model and project. | AI steps consume tasks and AI credits on every run. |
| InTouch AI | Zapier |
| Where your credentials live | An AES-256 vault on your machine. The assistant can USE a credential to do work without the secret ever entering its context. | Stored in Zapier’s cloud. Your bank, CRM and mailbox credentials sit on their servers. |
| Whether your data leaves the building | It does not have to. Run every step locally, with a local model if you want, and nothing crosses the network. | Every task runs in their cloud, so your data crosses it by definition. |
| Audit trail | Every action attributed and timestamped; every run keeps inputs, outputs and result code — in the free edition, not a tier. Both are readable by the assistant. | Task history for runs; account audit logs on higher plans. |
| Granting access to individual capabilities | Capability grants: allow or deny individual tools, AI providers, skills — and the assistant itself — per user or per group. Deny-by-default, enforced where a user reaches for the capability: listing tools, adding one to a workflow, running a skill, opening the assistant. (Enterprise; Department governs objects through roles.) | Shared connections and app restrictions on higher team plans. |
| InTouch AI | Zapier |
| What it takes to install | One JAR with an embedded database. About five minutes. A Raspberry Pi 5 is enough. | Nothing to install — a genuine advantage. |
| If the vendor disappears tomorrow | It keeps running. It is on your hardware, and the automation is JSON you already have. | Your automation stops. It only exists inside their service. |
| Air-gapped or regulated environments | Supported. No outbound dependency, and a fully local model if AI is required. | Not possible. |