01
Close is a scavenger hunt, not a calculation
The invoice is in a mailbox, the receipt is a scan in a shared drive, and the figure that settles the question is a row in a warehouse table. Nobody disputes the numbers. Finding them is the work.
FINANCE & ACCOUNTING
Promptev builds AI agents for finance and accounting teams that pull figures out of invoices, spreadsheets and the database holding the system of record, answer with the document behind each number, and take the next step — writing a sheet, opening an issue, posting a summary — only where you allow it.
1,000 free credits for a 30-day trial · no card · model calls go to your provider on your own keys
What gets in the way
The invoice is in a mailbox, the receipt is a scan in a shared drive, and the figure that settles the question is a row in a warehouse table. Nobody disputes the numbers. Finding them is the work.
The people who need the answer do not write queries, and the people who write queries have their own backlog. A database tool lets an agent query PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, ClickHouse or Snowflake in plain language — read-only by default, over only the schemas you make visible.
When someone asks for the contract or the receipt behind a figure, a confident summary is worthless. Every Promptev answer carries the documents it came from, and every run an agent has ever done is in the audit trail.
Worked examples
Use case 01
An agent reads the invoices arriving in Gmail or Outlook and the scans sitting in Drive, OneDrive or Dropbox, pulls out amounts, dates, suppliers and line items, and appends them to a spreadsheet. Scanned pages and photographs are read by a vision model on your own key.
How it holds upEmbedding your content is free; reading a document costs credits by its size, and pages that need a vision model cost more. Sensitive values can be masked before anything is stored.
Use case 02
Attach a database tool and an agent answers “what did we spend with this supplier last quarter” from the system of record itself, returning a table or a chart rather than a paragraph. Connections are read-only by default, and you choose which schemas and tables are even visible.
How it holds upSeven engines: PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, ClickHouse and Snowflake — under the same approval rules and the same audit trail as every other tool.
Use case 03
Run an agent every Monday morning across the sheets and sources you name, and have it post the summary into a channel or send it by e-mail. Several agents can be composed into an orchestration where each does a part and passes the work along, and a run that needs a person pauses and resumes where it left off.
How it holds upA run starts on a schedule, on a webhook, or when another agent calls it as a tool. A held run notifies the people who can release it.
Use case 04
When an auditor asks for the policy, the contract or the receipt behind a number, ask the agent instead of the shared drive. It searches everything the person asking is entitled to see and names the document, the way a working-paper reference does.
How it holds upAnswers follow current permissions, and the audit trail records every change and every run in language a person can read.
Beyond chat
Write what the agent should do in plain words, attach the tools it may use, and choose what starts it. Every tool that writes, sends or spends carries its own approval rule, and a run that stops for a person resumes where it left off.
Schedule — weeklyAn example run
How it works
Connect one source, wait for it to be read, and ask a question. An afternoon is a realistic first result, not a quarter.
Step 01
Authorise it once with your own account, then pick exactly what Promptev should read — a specific Drive folder, a Confluence space, a Jira project — rather than everything you can see. Permissions travel with the content, and it stays in step as files change.
Step 02
Say what it should do in plain words, attach the knowledge it may use, choose the model — on your own provider key — and give each tool its approval rule. Test it in the built-in chat before anyone else sees it.
Step 03
The App, Slack, WhatsApp, a widget on your own site, your code through a scoped API key, or any MCP-compatible developer tool. Roles decide who may build and who may only talk to it.
What it connects to
Eight apps to connect — Google Workspace, Microsoft 365, Dropbox, Confluence, Jira, Azure DevOps, Slack and WhatsApp — are 16 services and 198 tools. Everything else is reached the same three ways: any REST API as an HTTP tool, seven database engines as database tools, or any MCP server you run. All three obey the same approval rules and land in the same audit trail.
What stays yours
Each of these is independent of the others, so you can bring one, several, or all of them. What remains is the platform work — connectors, agents, approvals, audit, channels — which is the part worth buying.
Managed by Promptev, inside your own cloud account in whatever region you choose, or fully on your own machines. The product is the same in all three.
Model calls go to your provider on your own keys, including a self-hosted or OpenAI-compatible endpoint. A workspace can run its documents, embeddings and retrieval on a database you own, and keep files in your own bucket.
SOC 2 Type II, with a data processing agreement and a service level agreement available and sub-processors published. Your content is never used to train a model, and every answer names the documents it came from.
Start here
Connect a mailbox, a drive or a read-only database, and ask something you could check by hand. 1,000 free credits for a 30-day trial, no card.
Plain answers
Yes. Files are read and made searchable including scanned pages and images, which are handled by a vision model — on your own key, like every other model call. Spreadsheets and tables are read as tables rather than as prose, and sensitive values can be masked before anything is stored.
Yes, through a database tool. Seven engines are supported — PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, ClickHouse and Snowflake — connections are read-only by default, and you choose which schemas and tables are even visible. Results come back as tables or charts.
An agent, or an orchestration of several, runs on a schedule you set or when a webhook fires. It reads the sources you attached, does the work, and delivers the result to a channel or by e-mail. If a step needs a person to approve it, the run pauses and resumes where it left off once they do.
Only if you allow it. Actions are the tools you attach, and each one carries its own approval rule: no approval, so it simply runs — right for reading; ask once, where the first call in a conversation stops for a person and the rest of that conversation runs without asking again; or always ask, where every call stops. Requests are answered in the App, by e-mail or in Slack, and they expire rather than waiting forever. Every change and every run is in the audit trail.
Managed by Promptev, inside your own cloud account in whatever region you choose, or fully on your own machines — the product is the same in all three. Beyond that, a workspace can point its documents, embeddings and retrieval at a database you own, keep files in your own storage bucket, and register your own connector applications so the consent screen carries your company’s name. Model calls always go to your provider on your own keys.
The same as for one person, because there are no seats and no plans. Promptev is pay as you go: a new workspace starts with 1,000 free credits for a 30-day trial and no card, then you buy credits when you need them from $5. Giving everyone a viewer role costs nothing until they use it. Model costs are billed by your own provider on your own keys. Enterprise is an annual licence for a custom deployment in your cloud or on your premises, with onboarding and support — talk to us at [email protected].