Privacy-first, context-aware AI
Your people are already asking AI about work. Today they paste your documents into tools you do not control. Promptev gives them something better: assistants that know your files, answer with sources, and do the work — where each person only ever sees what they could open anyway. On your models, your data, your infrastructure.
2,500 free credits · no card · your own model keys
Why this is urgent
Nobody is waiting for permission. People are pasting contracts, board packs and customer data into whatever assistant is open in the next tab, because it saves them an hour and the alternative is asking a colleague who is in a meeting.
Banning it does not work; it moves it out of sight. The only thing that works is giving people something better, on your side of the fence.
How many days of parental leave do I get, and who signs it off?
Twenty weeks at full pay, after twelve months’ service. Your line manager approves it, and People Ops is notified automatically.Parental leave policy 2026 · §3Employee handbook · p.18
Can you start the request for October?
create_request · waiting for your manager’s approval
What it does, in three steps
No migration and no new place to log into. Promptev reads the systems you already pay for, and puts the assistants where your team already talks.
A Drive folder, a SharePoint site, a Confluence space, a Jira project, a Dropbox, a database, or files you upload. Whatever a person could open, it can read — including scanned pages and spreadsheets.
In plain words: “answer HR questions from the handbook”, or “review this month’s invoices and flag anything odd”. Give it the knowledge it may use, and set each tool to one of three: no approval, ask once in a conversation, or always ask.
In the browser, in Slack, in WhatsApp, in a chat box on your website, or inside your own software. Answers come back with the documents they came from, so anyone can check them.
And the part that makes it safe to switch on. Everyone asks the same assistant, but nobody gets an answer from a document they could not have opened themselves. That is decided as the question is answered, not by tidying up afterwards — which is why finance figures stay with finance while everyone else still gets a useful reply.
And the list does not end at sixteen
Eight apps arrive connected — 16 services and 198 tools between them, because Google Workspace is Drive, Docs, Sheets, Slides, Gmail and Calendar, not one connector. Everything else reaches Promptev by one of three routes, and none of them waits on us to build anything.
Point Promptev at any REST or HTTP API — an internal service, a partner endpoint, a product with no ready-made connector — describe its inputs, and it becomes a tool an agent calls under the same approval rules and the same audit trail as every built-in one. Not second-class.
Agents query your databases in plain language and hand back tables or charts. Read-only by default, and you choose which schemas and tables are even visible.
Attach any Model Context Protocol server you run or trust and its tools appear beside the built-in ones. It works the other way too: each project is itself an MCP server at its own URL, so any MCP-speaking developer tool gets that project's knowledge and tools with the caller's permissions intact.
The difference in ten seconds
Here is that promise, in practice: one assistant, one question, two people with different access. Anything that retrieves first and filters afterwards has already read the document. Promptev decides while it looks, so the wrong document is never fetched.
What did we pay Northwind in Q2, and is the renewal signed?
You paid Northwind $184,200 across four invoices in Q2. The renewal was countersigned on 12 June at a 6% uplift, effective 1 July.
What did we pay Northwind in Q2, and is the renewal signed?
I can’t see finance records for Northwind. From the project wiki: the integration is live and the renewal was announced as signed on 12 June. For amounts, ask the finance team.
Why this matters. Anything retrieved and then filtered has already been read, and it leaks into “related” answers, summaries and follow-ups. Here the finance files are never fetched for Tom, and the agent says so instead of guessing.
The context engine
Every agent on Promptev reads through one engine. It turns documents, scans, spreadsheets and messages into searchable knowledge, keeps who-may-see-what attached to every piece, and hands the model only what the person asking could open themselves — with the sources to prove it.
The Promptev App
Promptev is a product your whole team opens, not only an API. Owners and admins set up the workspace. Editors build agents. Viewers simply talk to them.
Reads the month's invoices, flags anomalies, drafts Jira tickets. Asks finance before filing.
Answers staff questions from the handbook and the policies each person is cleared to read.
Every Monday 07:00: compiles the week's case notes and posts the report to #finance.
Looks up a supplier across the ERP database and the contracts. Read-only, nothing to approve.
Embedded on the pricing and docs pages. Grounded in product docs and the customer's own account.
A team of three agents on the canvas: reconcile, review, report. Pauses for approval before posting.
How many days of parental leave do contractors get, and who approves it?
search_knowledge_base · 3 passages · as [email protected]
Contractors are not covered by the parental leave policy. Leave for contractors is agreed in the individual contract and approved by the engaging manager. Employees get 20 weeks; that section is not visible to you.Contractor handbook.pdf · p.4Leave policy 2026 · §2
Draft an e-mail to my manager asking for two weeks in October.
send_email · needs your approval before sending
Every agent in a project, who built it, where it answers, live or draft.
Files, voice notes and images in; answers with the documents they came from.
Connect a source or upload. Per-file sync, live progress, OCR for scans.
Connector tools, HTTP tools, database tools and your own MCP servers.
Teams of agents on a canvas. Schedules, webhooks, retries, pause for approval.
Approve in the App, by e-mail or in Slack. Expires if nobody answers.
Owner, admin, editor, viewer, plus project roles. SSO groups become access rules.
Every change and every run, who did it, readable by a person, exportable.
Slack, WhatsApp, your website, a portal, the API. One agent, every door.
Approvals, held runs, connectors needing attention, credits. In the App and by e-mail.
Pay as you go, auto top-up, receipts, invoices when you need them.
Scoped keys per project. A widget on your site with one script tag.
Inside a workspace
A workspace is the wall around your data. Inside it, projects hold knowledge, agents and tools. Documents flow in from the places they already live, pass the permission gate as each person asks, and answers flow out to wherever the team is.
Eight apps — Google Workspace, Microsoft 365, Dropbox, Confluence, Jira, Azure DevOps, Slack, WhatsApp — which are sixteen services and 198 tools once the sub-services are counted. Plus anything you upload.
A project holds several knowledge bases, each with its own sources. Text, tables and layout are extracted, scans are read by a vision model, and every piece keeps the permissions it arrived with.
The caller's identity and groups are part of the database query, so a document they could not open is never fetched in the first place.
One agent, or a team of them on a canvas, running on a schedule or a webhook — stopping for a person before anything that matters. HTTP, database and MCP tools attach here: they give an agent something to call, never a corpus to read.
The Promptev App, Slack, WhatsApp, a chat box on your site, a branded portal, your own code over the API or MCP.
Integrations
Eight apps to connect — Google Workspace, Microsoft 365, Dropbox, Confluence, Jira, Azure DevOps, Slack and WhatsApp. They are sixteen services once the sub-services are counted, because Google Workspace is Drive, Docs, Sheets, Slides, Gmail and Calendar rather than one connector. Everything else your company runs reaches Promptev by one of three open routes.
Sixteen ready, and an open edge for the rest: HTTP tools, database tools and your own MCP servers reach anything with an interface, under the same approval rules and the same audit trail as a built-in one.
Google API Services · Limited Use
Promptev connects to your Google Drive, Gmail, Google Calendar, and Google Sheets via OAuth. Only files and folders you explicitly select are indexed so your AI agents can search and reference them. Every tool action can require human approval before execution. Promptev is bring-your-own-key — your data is processed by the model provider you choose with your own API keys and is never used by Promptev to train any model. You can revoke access at any time from your Google account settings.
What teams build
Agents read, decide and act through the same tools your people use, and stop to ask before anything that costs money or leaves the building.
Reads the month’s invoices, flags anomalies, drafts the Jira tickets, asks finance before filing.
A team of three agents on a canvas: reconcile, review, report. Pauses for a person before anything posts.
“What did we spend with Northwind last quarter?” answered from the ledger database, as a table or a chart.
Reads an incoming contract against your playbook and flags the clauses that differ from your standard.
Answers staff questions from the handbook and the policies each person is actually cleared to read.
Pulls a supplier’s documents together and drafts the review pack from your template.
When an alert fires, gathers the runbook, the recent changes and the open tickets into one Slack message.
Turns the week’s merged work into notes a customer can read, posted where the team already is.
Your project’s knowledge and tools inside any MCP client, with the caller’s permissions intact.
On your site, grounded in the product docs and each customer’s own account data, nothing else.
Reads an incoming ticket, finds the matching history and drafts the reply for a human to send.
Summarises everything known about an account before the call, in one page.
New hires ask in WhatsApp; the agent answers from the handbook and the policies they are cleared for.
Finds a slot everyone can make and sends the invitation, once a person approves it.
Everything the team knows about an account — documents, mail, the CRM — gathered before the meeting.
Every Monday at 07:00, compiles the week’s numbers and posts the report to the team channel.
None of these is a template. They are the same agent builder with different instructions, different knowledge and different tools — which is why anything not listed here is a configuration task rather than a feature request: any REST API becomes a tool, any of eight database engines can be queried, and any MCP server can be attached.
Privacy first, in the literal sense
Bring your own models and keys, your own storage, your own database, your own connector apps. Run it managed by us, in your cloud, or fully on your premises.
Governance
Roles decide who may build, who may only talk to agents, and who approves. Every action an agent takes lands in an audit trail a person can read. Nothing that spends money or writes to your systems happens without the rule you set.
None of this is a feature you switch on. Roles, the audit trail and the approval rules belong to the workspace — every agent in every project runs inside them.
Where Promptev sits
Enterprise search, a suite copilot, a developer framework, or a pipeline you build yourself. Each is good at something. Promptev is the context engine and agent platform that retrieves as the person asking, acts through governed tools, and runs where you decide.
Glean is a serious product — hundreds of connectors, many models, and permission-aware access built in. Promptev differs on pricing shape and on where it runs: no seats and no minimum, and a workspace can run on its own database, its own models and its own storage, with the engine that enforces the permission boundary open for you to read.
Excellent inside Microsoft 365. Promptev spans Google Workspace, Microsoft 365, Dropbox, Confluence, Jira, Azure DevOps and your databases, and answers in Slack, WhatsApp, your own site and MCP clients. If you live entirely inside Microsoft 365, Copilot is a fine start and Promptev can sit beside it.
Frameworks give developers the parts to build a pipeline. Promptev ships the finished platform: connectors, a permission-aware context engine, agents, approvals and audit. Use them together through MCP — a project URL is an MCP server.
You own every line, and every incident. The context engine is open source under Apache-2.0, so take the engine alone or the whole platform — and keep the freedom to leave.
For builders
Promptev is an MCP server, a REST API and an open engine. Point Claude Code, Cursor or your own agent at a project and it gets that project's permission-aware knowledge and its governed tools, exactly as your people do — no more.
Orchestrate teams of agents that run on a schedule or a webhook, pause for approval, and pick up where they left off. Or skip the platform entirely and call the engine as a library: two clients, either language, one Postgres schema between them.
{
"mcpServers": {
"promptev": {
"url": "https://app.promptev.ai/mcp/acme/ops",
"headers": { "Authorization": "Bearer pk_…" }
}
}
}
// one server per project (acme / ops)
// search_knowledge_base · call_tool · call_agent — scoped to the key's grantsPricing
Pay as you go
Buy credits when you need them, use your own model keys, and pay only for what your agents do. Enterprise is an annual licence with a custom deployment — in your own cloud account or on your own premises — plus single sign-on, audit exports and support, sized to the estate rather than to a seat count.
Plain answers
The part of an AI system that turns your documents, spreadsheets, scans and messages into retrievable knowledge and decides what the model is allowed to see for each request. Promptev’s context engine is open source under Apache-2.0 and carries the caller’s permissions into the retrieval query itself.
Retrieval where the person’s identity and groups are part of the database query, so documents they cannot open are never fetched, rather than fetched and filtered afterwards. It is the difference between an AI you keep to one team and one you can put in front of contractors.
A chatbot answers. An agent reads, decides and acts through tools: it files the ticket, creates the folder, sends the report, on a schedule or when asked. On Promptev every tool is set to one of three — no approval, ask once (the first call in a conversation stops for a person, then the rest of that conversation runs), or always ask — and every action lands in an audit trail.
Sixteen services across eight apps arrive ready, with 198 tools between them. Beyond those there are three open routes, and none of them waits on us. HTTP tools: point Promptev at any REST or HTTP API, describe its inputs, and it becomes a tool an agent calls under the same approval rules and the same audit trail as a built-in one. Database tools: agents query your databases in plain language and hand back tables or charts — PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, SQLite, ClickHouse and Snowflake — read-only by default, and you choose which schemas and tables are even visible. Custom MCP servers: attach any Model Context Protocol server you run or trust and its tools appear beside the built-in ones. It works both ways: each project is itself an MCP server at its own URL, so any MCP-speaking developer tool gets that project’s knowledge and tools with the caller’s permissions intact.
Glean is a serious product — hundreds of connectors, many models, and permission-aware access built in. Promptev differs on pricing shape and on where it runs: no seats and no minimum, and a workspace can run on its own database, its own models and its own storage, with the engine that enforces the permission boundary open for you to read. If you live entirely inside Microsoft 365, Copilot is a fine start and Promptev can sit beside it.
Yes. Every project is an MCP server: one URL gives any MCP client that project’s permission-aware knowledge and governed tools, with the caller’s permissions intact. The context engine is also a library you can call directly — pip install promptev-context-engine for Python, npm install @promptev/context-engine for TypeScript — and both clients share one Postgres schema, so a corpus ingested by either is searchable by the other.
Managed by Promptev, in your own cloud, or fully on-premises. Model calls go to your provider on your own keys — chat, embedding and OCR alike — so switching model or provider does not change what you built. Files, embeddings and the knowledge graph live in a database and a bucket you can point anywhere. SOC 2 Type II.
Pay as you go from $5, with 2,500 free credits and no card to start. There are no seats: you pay for what your agents do, on your own model keys. Enterprise is an annual licence with a custom deployment in your own cloud account or on your own premises, plus single sign-on, audit exports and support — sized to the estate, not to a seat count.
Talk to a person
A question about running it in your own cloud, a connector you need, a security review to get through — write it here and it reaches the team directly, not a queue.
Start here
Two thousand five hundred free credits, no card. Or talk to us about running it where your data already lives.