Start with the outcome.
Learn the system as you go.
Invice is the agentic operating system for client-facing work, a system of action over the systems of record. You do not write code to shape it. You compose it. This guide is built as a ladder: pick an outcome and build it in minutes up here, understand the pieces in the middle, and go as deep as the reference at the bottom whenever you need it.
What do you want Invice to handle?
You did not sign up to learn our nouns. You signed up because something keeps slipping: a lead going quiet, a document never sent back, a renewal creeping up unwatched. Start from the thing you want off your plate. The concepts will introduce themselves along the way.
Follow up with silent leads
Keep reaching out until they reply or book, with the cadence and the stop rule you set.
Collect missing documents
Ask, remind, track what is still outstanding, and escalate to you when a file stalls.
Keep every client moving
Watch for silence, deadlines, and stalled next steps across your whole book.
Coordinate an onboarding
Run one repeatable process across every new client, each with its own mission.
Put an agent on my website
Answer visitors from your approved documents only, verify them, and capture interest.
Connect my existing stack
Work through Gmail, Outlook, your CRM, and nearly any tool with an API.
Your first mission, start to sent.
One connected inbox, one clear goal, one approved send. About five minutes, and every screen named below is the real one. Nothing reaches a client until you approve it, so there is no way to break something by trying.
A faithful still of Step 1 in the real mission wizard: the goal in plain language, and the Stack and Vault rows underneath it scoping the apps this mission may act through and the documents that ground it.
Connect a model and your inbox
Invice runs on your own AI model key, so connect a model first. Then connect Gmail or Outlook. The agent sends from your real address and watches that same inbox for replies.
Open the mission wizard
Four steps: Define goal, Clarify, Configure, Launch. You will spend nearly all of your time on the first one, and that is the point.
Say what done looks like
Plain language, outcome first. For a silent lead: follow up until they reply or book a 30 minute intro call, using the pitch in the vault. Be specific about the what, quiet about the how.
Add the limits in the same breath
Every line the agent must never cross goes in the brief: never quote a price, wait at least three days between touches, stop after three attempts. Corrections you make later are saved as new rules automatically.
Answer what the agent asks
Invice reads your goal and asks the few questions a good assistant would: timeline, budget, constraints. Your answers are folded back into the brief before anything runs.
Pick what it can see and use
Tick the vault documents it may share, each marked with approval, and confirm the connections this mission is allowed to act through.
Launch, then approve the first send
The agent plans the steps and drafts the first email, and the draft waits for you. Review it in your queue, edit it if you like, approve it. Nothing reaches a client until you do.
Where step 7 ends: the drafted send waiting in your approval queue.
The minimum a workflow needs.
Strip any Invice workflow to the studs and four things remain. If you can answer these four questions about a piece of work, you can build it here, whatever the industry.
Outcome
What should happen? One sentence, outcome first. In Invice this is the mission’s Goal, and it is the only part you cannot skip.
Context
What should the agent know? A vault document, the email thread, the client’s fields. Grounded action means it checks the source before it acts, and escalates instead of guessing.
Authority
What may it do? The connections it can act through, the guardrails it must never cross, and your approval on every outside write by default.
Stop condition
When is it done? Until they reply, until a date, or after a set number of tries. Without one, a reminder never ends.
The learning path.
Nothing on this page is required on day one. This is the order that works. Each tier links to its chapter below, and everything past the first column can wait until the simple version is running.
- Connect a model and your inboxThe only setup a first mission needs.
- Create one missionThe wizard walks you through it.
- Write a goal with limitsOutcome, guardrails, stop condition.
- Approve each actionEvery outside write waits for you.
- Let Triggers start the workDeadlines, silence, schedules.
- Wrap a process in an OperationOne parent, a mission per client.
- Line up fields in the Data mapWhen a CRM enters the picture.
- Save your setup as a PlaybookRerun it in one click.
- Connect anything via MCP or RESTBeyond the built-in paths.
- Put the Gateway on your siteA governed agent, one embed tag.
- Partner agents with A2AStructured asks between workspaces.
- Opt tools into automatic runsOnly the ones you choose.
Six proven patterns.
The building blocks combine into an endless range of workflows. Here are six proven shapes to build from, or fork whole from the Marketplace. Each names the pieces it leans on and how far up the learning path you should be before reaching for it.
Silent-client re-engage
When a client has not been in touch for a number of days, the agent opens a warm check-in and books time, or escalates if the file looks at risk.
Lease-expiry reminders
A trigger fires at 90, 60, and 30 days before a lease ends, each with firmer copy, until the tenant replies.
Renewal operation
An operation that runs a renewal across your whole book. A child mission per client, briefed from the vault, escalating only the rate decision.
Front-door concierge
The Agent Gateway answers site visitors from approved docs, verifies them, and hands qualified interest to your triage.
Newsletter re-engagement
Describe an audience in plain language, the agent drafts and personalizes from your latest content, and waits for your send.
Referral loop
A partner’s agent asks yours whether a lead fits your criteria. The ask lands in triage, you approve, your agent answers.
You compose Invice. You don’t code it.
You have already met the four ingredients: outcome, context, authority, stop condition. Here is the same model in Invice’s own nouns. Every workflow is built from the same small set of parts. You brief a Mission with a Goal in plain language and the Guardrails it must never cross. You choose which Blocks that mission shows, connect the outside systems it should act through, and the agent plans and runs the steps between, returning to you only for the decisions that are genuinely yours.
For a process you repeat across many clients, you wrap it in an Operation. To make work start on its own, you add Triggers. Once a mission is shaped the way you like, you save it as a Playbook and reuse it, or publish it to the Marketplace for other professionals to run. That is the whole model. The rest of this guide is depth on each part.
Vocabulary
A display skin over the same data. Keep Invice’s business language, or switch to the terms a Salesforce or HubSpot team already knows. Nothing about your data moves.
Blocks
The panels a mission shows. Add the ones the work needs, drop the ones it does not. One of them is the Data map, where your fields line up with a connected system.
Goal & guardrails
The brief in two parts. The goal is the outcome you want. The guardrails are the lines the agent will not cross. Write both in plain language.
Playbook
A mission, with its blocks and its rules, saved as one template. Run it again for the next client, or publish it to the Marketplace.
Missions
A Mission is one job the agent owns end to end, usually for one client. You brief it like a teammate, and it works the job from first contact to done, proposing actions and waiting for your approval by default.
›The lifecycle
The four ingredients you write are the input. The lifecycle is what the agent does with them, and every mission moves through the same six phases. You are only pulled in at Escalate.
›The four shapes
Most missions are a variation on one of four shapes. Naming the shape helps you write a tighter brief.
Monitor
Watch for a condition and act when it is met. A lease nearing its end, a client going quiet, an inbox that needs triage.
Search
Go find something and bring it back. Pull a document, gather answers, assemble the facts a decision needs.
Coordinate
Move a multi-step process forward across people and systems. Onboarding, renewals, collecting a file.
Remind
Reach out on a cadence until a thing happens, escalating the tone politely as time runs down.
Operations
An Operation is the parent that runs one process across your whole book of clients. It never executes on its own. It holds the shared brief, a goal template with placeholders like [CLIENT NAME], the pipeline stages, the connections, the vault documents, and the rules. When a new client arrives, by hand or by inbound email, the operation spawns a child mission that inherits all of it.
Persistence
An operation is always on. It does not end when one client finishes. It keeps running your book.
Composition
Set the brief, blocks, and rules once on the parent. Every child mission inherits them automatically.
Discovery
Route inbound mail to the right operation, so a new client can start a mission without you lifting a finger.
Memory
What the agent learns on one client informs how it carries the next, across the whole operation.
Blocks
Every mission has a rail of Blocks, the panels that make up its workspace. A recruiting mission and a mortgage file need different surfaces, so you add the blocks the work needs and drop the ones it does not. You choose once, per mission or per workspace, and it holds.
| Block | What it shows |
|---|---|
| Brief | The goal, the inputs, and where the work stands right now. |
| Communication | Every email in and out of the mission, in one thread. |
| Documents | The files this mission collects, sends, and reads from. |
| Requests | Open asks and what each one is waiting on. |
| Property | For real-estate and lending work: the property, unit, or loan at the center of the file. |
| Timeline | The chronological record of every step the agent took. |
| Data map | How your fields line up with a connected system, and who wins when they differ. |
| CRM | The relationship view: priority, notes, planned outreach, reminders. |
Adding a block is one choice on the mission’s right rail. The set you pick is part of what a Playbook captures, so a shared workflow arrives with the right surfaces already on.
Goal & guardrails
The brief is two parts. The goal is the outcome you want, written like a note to a teammate. The guardrails are the lines the agent will not cross, kept in their own block so they hold on every single action rather than getting buried in the goal.
Book a 30 minute intro call with each new lead this week, using the pitch in the vault. Be specific about the what, and quiet about the how.
GuardrailsWriting a good brief
The agent builds the steps. Your job is to be clear about the outcome and the limits. A few habits make the difference between a mission that runs itself and one that keeps asking.
›Name the outcome, not the steps
Say what done looks like and point to the inputs. Be specific about the what, quiet about the how. If you can brief a person, you can brief a mission.
›Put every limit in a guardrail
Anything the agent must never do belongs in the guardrails block, not the goal. Rules there hold on every action. Goals drift.
›Say when to stop
For anything that repeats, name the stop condition: until they reply, until a date, or after a set number of tries. Without one, a reminder never ends.
Triggers
A Trigger is a rule that starts work on its own: on a schedule, on a time-based condition, or on an inbound event. You write it in plain language and Invice turns it into a precise, structured rule. When it fires it either flags you, launches a mission from a goal template, or calls a connected tool.
Time-relative
A lease expiring in 90, 60, then 30 days. A client silent for two weeks. No reply after your outreach. It fires once per stage, with escalating copy.
Scheduled
Once at a set time, or recurring by day, week, or month in your own timezone.
Inbound webhook
An outside system pushes an event, and an optional filter decides whether it fires. A failed payment, a new signup.
Three actions
Notify you, launch a mission from a template, or call a connected tool. Every write still passes your approval.
Connection methods
Invice acts through the systems you already use. There is no fixed integrations list to wait on. Instead there are a handful of open ways in, and almost every tool fits one of them. Here is what connects today, and what is on the way.
Email & calendar
Live todayDirect sign-in. Your agent sends and reads from your own inbox, and schedules on your own calendar.
MCP servers
Live todayAny app that speaks the Model Context Protocol. Paste a token and its actions become tools your agent can call.
MCP servers, OAuth sign-in
Coming soonThese have a hosted server but need one-click sign-in, unlocked in an upcoming release. Some can be reached today through a REST recipe or self-hosted server when that path fits; otherwise they wait for OAuth sign-in.
Guided REST builder
Live todayPoint the builder at a documented JSON REST API, add bearer/API-key auth, declare endpoints, and test before tools activate. One-click recipes pre-fill common apps.
Other ways in
Live todayThe Data map
The Data map is the block where you line up an Invice field with a field in a connected system, and name the source of truth. Your Email to Salesforce Contact.Email, your Stage to its Opportunity stage. For each pair you say who wins when the two differ, and whether Invice reads it, writes it, or both.
| Invice field | Connected field | Source of truth | Direction |
|---|---|---|---|
| Contact.Email | Invice | Both | |
| Stage | Opportunity.Stage | Remote | Read |
| Phone | Contact.Phone | Most recent | Both |
Today the map is declared connection context the agent can use when reasoning about fields, source-of-truth rules, and connector schemas. Live scheduled two-way sync and universal write enforcement are on the roadmap.
The safety model
Openness only works if it is safe by default. Three rules hold across every connection.
Writes wait for you
Any tool that changes something in an outside system requires your approval when the agent is running unattended. You opt individual tools into automatic only when you choose to.
Keys never travel
Every credential, token, and key is encrypted at rest with AES-256, decrypted only in memory for an active run, and never placed into the model’s context.
Untrusted by default
Inbound messages and partner asks are treated as untrusted input. They land in triage, never in an auto-executing path.
Agent Gateway
The Agent Gateway puts one governed agent on your own website with a single embed tag. Visitors chat with it. It answers only from the vault documents you switch on, proves who is real on both sides, captures interest the moment it appears, and hands off to a person when asked, with a subtle Invice Gateway Agent trust label.
Grounded, or not at all
Only from the documents you approve. It does not invent an answer, and it says plainly when something is outside what it knows.
Both ways
Your verified profile lets a visitor confirm the real you. A one-time email check lets you confirm a visitor before anyone follows up.
No lead left on the page
When someone is ready, it takes their details and hands the conversation to your team. Every request lands in your workspace.
A person, one step away
It escalates rather than guessing. Governed actions on your connected apps are coming soon, always behind your approval.
Agent-to-agent
Two Invice professionals can partner their agents. A partner’s agent sends yours a structured ask, a question, and it lands in your triage. Nothing runs on its own. You approve every answer. Neither side can read the other’s data, and either side can cut the connection instantly.
Each workspace publishes a public agent card, a business card for its agent. Full end-to-end negotiation, where both agents settle a question between them, is on the roadmap.
Playbooks
A Playbook is a mission or operation, with its goal, its blocks, and its rules, saved as one reusable template. Capture is always private first. You publish only when you choose to, and client details are stripped before anything goes public.
Private by default
Saving grabs the goal template, the block set, and the active rules. It never leaves your workspace at this step.
Redacted, then public
Before a Playbook goes public, a deterministic pass scrubs names, emails, phones, and links. If it cannot verify the scrub, it refuses to publish.
A clean copy
Anyone can fork a public Playbook into their own workspace as a fresh private copy. Later changes to the original do not reach back into theirs.
Your own library
Keep private Playbooks to rerun your own setups across clients, no publishing required.
The Marketplace
The best Playbooks come from people doing the real work. The Marketplace is where they live: proven client-facing workflows, published by other professionals, ready to browse, fork, and run in your own workspace. Build one that saves you an hour a day, publish it, and it saves the next professional the same hour.
By the work, not the tool
Each listing shows the kind of workflow, the apps it runs on, and who built it, so you know what you are bringing in.
Reviewed and reportable
Comments, author attribution, and a report path keep the shelf honest. Client data never rides along.
Vocabulary
Invice speaks its own business language by default. If your team lives in a CRM, switch to CRM standard in Settings and every label becomes the term you know. It is a display skin only, applied per workspace. Nothing about your data changes.
| Business language | CRM standard | What it means |
|---|---|---|
| Client | Contact | An individual person |
| Company | Account | An organization |
| Goal | Goal | What the agent is asked to achieve |
| Deal | Opportunity | A qualified sales opportunity |
| Stage | Stage | A step in the pipeline |
| Timeline | Activity | Chronological history of interactions |
| Decision | Approval | A moment that needs your sign-off |
| Guardrail | Rule | A business rule the agent cannot break |
| Playbook | Playbook | A reusable, installable process |
| Request | Case | An inbound support request |
| Memory | Notes | What the agent remembers about a client |
Connections
| Method | What it connects | Status |
|---|---|---|
| OAuth | Gmail, Outlook, Google Calendar | Live |
| MCP server | Token-ready Model Context Protocol apps; GitHub, Linear, Stripe, Zapier, Fireflies, Supabase and more | Live |
| MCP, OAuth sign-in | Notion, Salesforce, HubSpot, Slack, Drive, DocuSign and more | Soon |
| Guided REST | Documented JSON REST APIs with bearer/API-key auth; recipes for HubSpot, Calendly, Airtable | Live |
| A2A | Another Invice workspace’s agent | Live |
| CRM webhook | Your CRM pushes contacts and events to a signed endpoint | Live |
| Import | CSV and file import with column mapping | Live |
Triggers
| Trigger | Fires when | Common action |
|---|---|---|
| Lease expiring | A set number of days before a lease end date, in stages | Launch a reminder mission |
| Client silent | No message from the client for a number of days | Launch a check-in mission |
| No reply after | No reply a number of days after your outreach | Launch a follow-up |
| Schedule | Once, or recurring by day, week, or month | Notify, or launch a mission |
| Inbound webhook | An outside event arrives and passes your filter | Call a tool, or launch a mission |
Stop conditions: until replied, until a date, or after a maximum number of tries.
Roadmap
What is shipped, and what is planned. Nothing here is aspirational dressed as done.
›Shipped
›Planned
From your workflow
to the whole community.
The best Playbooks come from people doing the real work. Build one that saves you an hour a day, publish it, and it saves the next professional the same hour. That is the network you are joining.