An AI that actually knowsyour agency
A Claude-powered assistant built into your workspace. It reads your own clients, projects and invoices to draft, summarize, surface and analyze — and never sends or creates anything without your say-so.
Read-only by default • You confirm every action
See it in action
Conversations that get real work done
Ask in plain language and watch the assistant read your workspace, think, and stream a useful answer — just like having a junior teammate on call.
What it is
A teammate that lives inside your workspace
Not a generic chatbot bolted on the side — an assistant that understands your agency because it's wired into the same data you work in every day.
The assistant is powered by Anthropic's Claude and connected to your own clients, projects, invoices and tasks. Ask a question and it quietly reads the relevant records, then streams back a grounded answer — a summary, an analysis, or a ready-to-edit draft.
Because it works from your real data and confirms before acting, it feels less like a gimmick and more like a sharp junior teammate who already knows the accounts.
What it can do
Five ways it earns its keep
From drafting to deep analysis, the assistant takes the busywork off your plate so your team can focus on the work clients actually pay for.
Draft
Follow-up emails, proposals and invoices written in your voice — ready to review and send.
Summarize
Condense a client account, a project's status, or a long thread into a clear, skimmable brief.
Surface
Spot overdue invoices and clients who've gone quiet before they slip through the cracks.
Analyze
Ask plain-language questions about revenue, margins and profitability and get real answers.
Automate
Hand off the repetitive busywork so your team spends its hours on billable, high-value work.
How it works
From question to confident answer
A simple, secure loop — you ask, it reads only your data, it responds, and you stay in control of every action.
Ask in plain language
Type a question or pick a suggested prompt — no commands, filters or query syntax to learn.
It reads only your data
Scoped, read-only tools fetch just the clients, projects, invoices and tasks in your workspace.
It drafts or answers
Claude streams a response in real time — a summary, an analysis, or a ready-to-edit draft.
You review & confirm
Nothing is sent or created until you say so. You stay in control of every action.
You ask
“Which invoices are overdue?”
Scoped tools run
- list_overdue_invoices
- get_client
- Read-only • your workspace only
You get
3 overdue invoices totaling $9,400 — oldest 21 days. Draft reminders?
How to use it
Up and running in four steps
No setup, no prompt engineering. If you can describe what you need, you can use the assistant.
- 1
Open the assistant
Click the Assistant in your workspace sidebar. It opens with your agency's context already loaded.
- 2
Pick a prompt or type your own
Start from a suggested prompt or ask anything in plain English — like talking to a teammate.
- 3
Review the response
Read the streamed answer or draft. Ask follow-ups to refine the tone, scope or details.
- 4
Apply it
Send the email, save the invoice, or copy the summary. The assistant only acts on confirmation.
Suggested prompts
Example prompts
Steal these to get started
Real things teams ask every day. Copy one, tweak it for your agency, and watch the assistant get to work.
“Draft a follow-up to clients I haven't contacted in 30 days.”
“Which projects lost money last quarter?”
“Summarize the Fabrikam project for our weekly review.”
“Which invoices are overdue, and by how much?”
“Draft an invoice for the Northwind retainer this month.”
“Which clients are at risk of going quiet?”
“What was our revenue and margin this quarter?”
“Write a proposal outline for the Contoso rebrand.”
“What are my open tasks due this week?”
By role
Something for everyone on the team
The same assistant, tuned to what each person on the team needs to move faster.
Owner
- Quarterly revenue, margin and profitability in plain language
- Early warning on overdue invoices and stale clients
- A live read on the health of the whole agency
Account manager
- One-line account summaries before every client call
- Drafted, personalized follow-ups for quiet relationships
- Instant context across projects, invoices and history
Project manager
- Status summaries for weekly reviews in seconds
- Budget and burn-rate questions answered instantly
- Open tasks and deadlines surfaced across projects
Finance
- Overdue invoices listed with amounts and ageing
- Drafted, on-brand payment reminders ready to send
- Natural-language revenue and profitability analytics
Across your agency
Wired into every part of the platform
Because the assistant shares the same data as the rest of Host Agency AI, it works fluently across every feature you already use.
Client management
Summarize any account, surface clients who've gone quiet, and draft personalized outreach.
Invoicing & payments
List overdue invoices, draft new ones, and write friendly payment reminders on demand.
Project & time tracking
Get project summaries, check budgets and burn rate, and pull your open tasks for the week.
Team collaboration
Condense long threads and updates into a clear brief so the whole team stays aligned.
Document management
Draft proposals and outlines, and summarize the contracts and deliverables in context.
Analytics & profitability
Ask plain-language questions about revenue, margins and which projects made — or lost — money.
Dashboards
Turn your live numbers into narrative answers — what's up, what's at risk, what needs attention.
Privacy & trust
Powerful — and built to be trusted
The assistant is helpful precisely because it's careful. Here's exactly how your data is handled.
Not used to train third-party models
Your prompts and workspace data are sent solely to generate your responses — never to train outside models.
Scoped to your organization
Tenant-scoped, read-only tools mean the assistant only ever sees data inside your own workspace.
Keys stay server-side
API access lives on our servers. The assistant reads through controlled, auditable tools only.
You stay in control
Human-in-the-loop by design. The assistant never writes or sends anything without your confirmation.
Read the full details in our Privacy Policy and Subprocessors.
Built on Claude
Your assistant runs on Anthropic's Claude — one of the most capable and safety-focused AI models available. It streams responses in real time and grounds every answer in your live workspace data, so you get fast, relevant help instead of generic guesses.
FAQ
Questions, answered
The things teams most want to know before they put the assistant to work.
Yes. The assistant uses tenant-scoped, read-only tools, so it only ever accesses data inside your own workspace. Prompts and the relevant workspace data are sent to the AI provider solely to generate your response — they are not used to train third-party models. API keys live server-side.
No. The assistant is read-only by default and human-in-the-loop by design. It can draft an email, propose an invoice, or surface an issue, but nothing is ever sent or created until you review and confirm it.
The AI assistant is built into the Host Agency AI workspace. See our pricing page for the plans it's included on, and start a free trial to try it with your own data.
Like any AI, it can occasionally get things wrong, so treat its drafts and answers as a strong first pass. Because it works from your real workspace data and you confirm every action, you always have the final say.
Put an AI teammate to work today
Start free and ask your first question in minutes. No credit card required — and you confirm every action.