It shows, instead of telling
Rather than describing five steps, the agent finds the right product tour and plays it — or composes a walkthrough on the spot, with coach-marks on your real interface.
Stept is the open-source Intercom + Fin alternative: a shared inbox across every channel, an AI agent grounded in your own knowledge base, a help center, product tours and automation. Self-host it free — bring your own model key — or let us run it.
MIT-licensed · runs offline with zero API keys · Postgres or SQLite
Connects to the tools your customers and team already use
Most support AI stops at text. Stept’s agent reads your docs, guides people through your actual UI, and — with consent — acts on the page. Every run leaves a full step-by-step trace you can audit.
Rather than describing five steps, the agent finds the right product tour and plays it — or composes a walkthrough on the spot, with coach-marks on your real interface.
With the visitor’s explicit opt-in it clicks, fills and navigates for them. Sensitive tools pause for human approval, password fields are off-limits, and actions are capped per run.
Claude, Cursor or any MCP client can search your knowledge, read conversations, write docs — even drive a real browser through the recorder extension.
Intercom bundles the inbox, the AI agent, the help center and product tours, then charges for each. Stept ships all of it in one open-source platform.
Pick the model, write the persona, and set every tool to auto, approval-required or off. Test it in a sandbox that shows the live reasoning trace before anything reaches a customer.
Crawl, upload or sync. Sources re-index on a schedule and prune pages that disappear.
Record a flow in your own app with the Chrome extension; steps self-heal when your markup moves.
Assign, snooze, tag, note and resolve together — with SLAs, saved views, macros and realtime collaboration across the team.
Crawl your site, upload files, or sync Notion, Confluence, Drive, Zendesk and GitHub. Hybrid retrieval feeds the agent, the reply copilot and your help center.
Rules, macros, campaigns, SLAs and webhooks — plus analytics on volume, resolution rate, CSAT and how often the AI deflects a conversation.
Multi-workspace tenancy, RBAC with custom roles, audit log, scoped API keys, encrypted provider secrets and a hardened deploy.
Most AI-support tools bill per resolution or lock the agent behind an enterprise plan. Stept ships the agent in the open-source core. Point it at your own model key and self-host the whole platform at no cost.
$ git clone https://github.com/myfoxit/stept-now stept
$ cd stept && make setup && make dev Early access: Stept Cloud is free right now — below is the pricing that’s coming.
Run the whole platform yourself.
Read the guideWe host, update and back it up.
Start freeFor teams with compliance needs.
Talk to usAI resolutions beyond your plan will bill at about $0.30–0.49 each — around half of Intercom Fin’s $0.99. A 5-seat team pays about $95/mo on Stept Cloud; the same team on Intercom + Fin typically runs $1,000+/mo.
Yes — the whole platform is MIT-licensed: inbox, AI agent, knowledge base, help center, tours, automation. Self-host everything with your own model key. Cloud is just us running the same code for you.
OpenAI, Anthropic, Google, Ollama, or any OpenAI-compatible endpoint. Keys are stored per workspace and encrypted at rest, and there is no markup on tokens — you pay your provider directly. A deterministic mock provider ships in the box, so the entire AI stack runs offline with zero API keys.
A conversation the agent resolves without a human stepping in. Self-hosted installs meter nothing at all. On Cloud, plan quotas apply once pricing turns on — during early access everything is free.
Only with the visitor’s explicit per-conversation consent. Actions are capped per run, password fields are never touched, navigation is same-origin by default, and any tool can be set to require a human approval before it executes.
Self-hosted: in your own Postgres, full stop. On Stept Cloud your workspace runs on EU infrastructure, model keys are encrypted at rest, and your conversations are never used to train anything.
Import contacts by CSV, pull help-center content in with the website crawler or the Zendesk connector, and keep your address by forwarding mail to your Stept inbox. Email us and we will help.
Start free on our cloud, or deploy it yourself with Docker Compose.