# Thanks, Computer ## Long-term goals. Everyday decisions. Thanks, Computer is a programmable runtime for durable, human-in-the-loop workflows. Pause an operation for a slow model, a webhook, or a human approver — for days — and resume exactly once, surviving restarts. Any language, plain JSON, every step replayable. Most workflow engines automate people *out* of a process; here a person — or a slow AI — is a first-class participant, not an exception. Bundle a set of operations into a reusable _stack_, and curate stacks into a **department** — a Support department, an Invoicing department, an Onboarding department — that runs a real business function on your own data, carrying its context across emails, web requests, schedules, approvals, and AI. > Your LLM has context. Why shouldn't your work? ### What people build with it: - a support department that drafts replies with AI and pauses for a human to approve - qualify sales opportunities against your ideal customer profile - prioritize customer success work around retention - guide partnership pipelines with strategic priorities - keep fundraising and business development efforts in context - give AI workflows the company context behind the task ## How it works Work is broken into small steps called **operations**, organized into an **op stack**. Each operation is gated by a [**resonator**](./resonators.md) — a trigger condition that fires only when the right kind of event passes by, and says what to do when it does: ```txcl WHEN @web.req.url.path == "/opportunity" EXEC "https://api.example.com/enrich-opportunity" ``` Everything at the same step runs in parallel; each operation's output deep-merges into the shared document, which carries the flow to the next step. ```stack opportunities 50 get_mission get_okrs 100 enrich 200 score review 250 assign 300 followup ``` Because the event is shared external memory — not parameters threaded through call chains — humans, AI, and services can collaborate on the same work: an AI operation drafts, a human operation approves, a service operation ships, all reading and writing the same context document. ## Composable by design You can think of operations as being black boxes. They receive JSON, emit JSON, and **can be written in any language**. Emails, web requests, AI tool calls, and schedules all become events flowing through the same shared document. A stack's operations may span many systems, but its behavior remains visible: - **Visible decisions:** Replay every flow step-by-step. - **The why travels with the work.** Context stays attached as work moves. - **Built for waiting.** [Continuations](./continuations.md) let an operation suspend — for a slow model, a webhook, a human reviewer — and resume days later, exactly once, surviving restarts. - **Any language, plain JSON.** Read JSON, write JSON. ## Run anywhere Self-host the open-source chassis on your own infrastructure, or let our cloud be the place to run it — managed reliability, email deliverability, and fleet scale, so your departments run like infrastructure instead of a script you babysit. The CLI `txco` targets either. The chassis is free and open source under the MPL-2.0 — self-host it, commercial use included. **Try it in two minutes:** [Quickstart](./quickstart.md) ```bash txco demo ```