> ## Documentation Index
> Fetch the complete documentation index at: https://docs.thread.app/llms.txt
> Use this file to discover all available pages before exploring further.
# Guides overview
> A guide is a resident agent you author in plain language that drives one named customer to one verified outcome over text, email, and voice.
A **guide** is a resident agent that lives on a customer relationship and drives it to one outcome — Profit Assist connected, the rollout completed, the add-on adopted. You write its instructions in plain language, give it channels to reach people on, and attach the tools it reads the world with. Then you enroll a customer, and the guide runs the conversation itself: it opens, waits, nudges, answers questions, and checks your system of record all the while.
The core promise is that a guide **never verifies its own work**. It closes an outcome only by citing a fresh read from a tool you attached — a HubSpot record, a Jira issue, a DocuSign envelope — or, when nothing attached can read the world, by proposing done and waiting for your tap. Neither the guide's word nor the customer's word closes anything.
## What a guide actually does
Between messages, a guide is asleep. It wakes on exactly three things:
* **An inbound reply** — the customer texts or emails back, and the guide answers in seconds, in context.
* **Its own timer** — the guide schedules its own check-ins ("give them until Friday, then nudge"), and each timer wake is logged with the guide's own reason.
* **You** — from the watch page you can **Message** the guide, **Nudge now**, or **Pause** it.
On every wake the guide decides what to do next: reply, wait longer, check the record, page you, or propose done. Every send, read, and decision lands in a per-customer log with receipts — see [the watch page](/guides/watch-page).
## Guide or Motion?
Guides and [Motions](/motions/overview) are the two ways Thread does work for you, and they compose.
| | A guide | A Motion |
| ------------ | ------------------------------------------------- | -------------------------------------------------------------------------- |
| Scope | One named person, one outcome | A whole account, a staged program |
| Structure | Plain-language instructions, no stages | Stages, plays, and milestone exit criteria |
| Conversation | The guide owns the thread — texting, email, calls | Plays run on a schedule and route through [workflows](/workflows/overview) |
| Done | One close, verified against your system of record | Milestones verified one by one as the account advances |
Reach for a guide when the job is "get this person to do this one thing, and prove it happened." Reach for a Motion when the job is a program — onboarding with phases, owners, and multiple verified milestones. A Motion can carry many kinds of plays; a guide is the resident line that works one human directly.
## The lifecycle
On the guide's **Design** tab, write the instructions — the outcome, the sentence that defines done, tone, cadence, and when to bring a human in. Attach tool rows, pick a texting number, and optionally set up email and phone calls. See [Building a guide](/guides/building-a-guide).
Before anything real, press **Try it live** and play the customer yourself — same instructions, same wake loop, shadowed sends, a virtual clock you can fast-forward. See [Rehearsal](/guides/rehearsal).
Publish's confirm step reads your instructions back verbatim alongside the close mode — *"closes on fresh evidence — reads: HubSpot"* or *"closes by your tap"* — so you see the definition of done and the evidence bound together at the moment they bind.
Enroll one customer by name, with a recorded consent attestation. Enrolling is the first wake — the guide opens the conversation itself; there is nothing to compose and nothing to schedule. See [Enrolling customers](/guides/enrolling-customers).
The guide texts, emails, or (with your per-call approval) calls — verifying against the source as it goes. When the customer claims they're done, the guide checks the record before celebrating.
The close cites the receipt that earned it. The Outcomes funnel shows every customer moving **enrolled → reached → engaged → verifying → done**, with verified closes distinguished from ones you confirmed by hand. See [Outcomes and verification](/guides/outcomes-and-verification).
## Guardrails are built in
Every guide carries a locked floor you cannot remove, enforced in code rather than in the prompt:
* **Consent before the first text** — no recorded consent, no send.
* **Quiet hours, recipient-local** — an overnight send waits for the recipient's open window.
* **STOP ends it, instantly** — a STOP reply suppresses that conversation permanently, with no model in the loop.
* **Never verifies its own work** — done closes only on a cited fresh read or your tap.
* **Pages you when it's stuck** — escalation to a human is never blocked or approval-gated.
You can add your own rules on top. The full floor is documented in [Compliance](/guides/compliance).
Guides is visible to **org admins only** — the same gate that covers Voice Studio, Cohorts, Motions, Agent Templates, and Workflows authoring. Other members supervise conversations from the [Inbox](/inbox/overview). See [User management](/admin-guide/user-management).
## Where to go next
Instructions, tools, channels, and publishing.
Single enrollment, consent, and what happens after.
Play the customer yourself before anyone real is enrolled.
The read, the funnel, and why green is earned.