Dhanur AI
Sign inGet started

For larger teams

Many agents, the same controls everywhere

When several teams run AI agents, you need one set of rules, a record of who did what, costs you can explain in rupees, and straight answers for your IT and legal teams. This is what Dhanur AI has for that today, and what isn't built yet.
Talk to us Governance checklist

  • Rules for every agent Live

    Settle things once for the workspace, such as “sending email always asks first” or “refunds over ₹2,000 wait for a person”. The stricter of a rule and an agent's own setting wins.

    How it works

  • Approvals, by action and by amount Live

    Any action can ask a person first. Actions with an amount wait when it's over your limit, and approvers can decide in bulk.

    How it works

  • Four roles Live

    Owner, Admin, Builder and Viewer, each limited to what the job needs. Contact details can be hidden from Viewers.

    How it works

  • Activity log Live

    Who changed what, by area, for Owners and Admins, with a CSV download.

    How it works

  • Tests before changes go live Live

    Test sets can block publishing until they pass, and daily checks score real conversations against your own rules.

    How it works

  • Everything that went wrong, in one list Live

    Failed tasks, connections to reconnect and messages that didn't arrive, each with one thing to do, and a daily email.

    How it works

The stack

One product instead of nine tools

Teams running AI agents often stitch together a separate tool for each job below, each with its own bill, sign-in and data terms. In Dhanur AI they are one product, billed in rupees, with its databases in India. Each row says what works today and what is still to come.

  • Start work when something happens

    Often bought as Zapier or Make

    LiveSchedules, webhooks, emails and chats, plus HubSpot contacts, new Google Sheet and Form rows, and Razorpay, Shopify and WooCommerce events. Agents can also set their own follow-ups.

    NextHundreds more apps as starting points.

    How it works

  • Know your business

    Often bought as Zep or a vector database

    LiveYour business details written once for every agent, files, website pages and tables, searched by meaning in English, Hindi and Hinglish, and long-term memory that asks before it saves.

    How it works

  • Connect to your apps

    Often bought as Composio

    LiveGoogle, Slack and HubSpot, your own tools with encrypted secrets, and any MCP server's tools. Dhanur AI is an MCP server too, so Claude Code or Cursor can build and run your agents.

    NextSign-in to MCP servers with OAuth.

    How it works

  • Pick the right AI model

    Often bought as OpenRouter

    LiveThree levels of Claude, a backup model when one is down, and your tests run on all three at once, so you can switch to the cheapest level that passes in one click. We don't route across AI companies: one set of data terms we can stand behind.

    How it works

  • Test before it goes live

    Often bought as Braintrust

    LiveTest sets with nine kinds of checks, including an AI judge, an option to block publishing until they pass, and daily checks on real conversations against your own rules.

    How it works

  • Build the agents

    Often bought as Dust

    LiveA builder with a test chat where nothing is really sent, a draft from a sentence, templates for Indian businesses, versions, and one-click undo to any earlier version.

    How it works

  • Make agents work together

    Often bought as CrewAI

    LiveWorkforces that hand work between agents, agents that ask each other and get a structured answer back, and several agents working on one message at once.

    NextRunning one agent for each item in a list.

    How it works

  • Never lose a job

    Often bought as Temporal

    LiveBackground work runs on a queue that keeps it through restarts and retries with growing waits. What still fails lands in one Needs attention list, with Run again.

    How it works

  • See what happened, and what it cost

    Often bought as Langfuse

    LiveEvery step of every task with its time and rupee cost, analytics by agent, model and tool, traces sent to your own monitoring tool, and a public uptime page.

    How it works

Tool names are examples of what teams buy for each job; we don't claim to match every feature of each. For approvals, roles, audit and where your data lives, see the governance checklist.

Before you decide

What we don't have yet

We are in early access. If any of these matters to you, tell us: it moves up the list.

  • PlannedMore seats and projectsEach workspace has 2 seats and 1 project today, with no monthly plan. Talk to us if you need more.
  • PlannedSingle sign-on and SCIMSign-in is by email code or Google today.
  • PlannedSOC 2We will not claim it until an independent audit is complete.
  • PartlyData in IndiaDatabases are in Bengaluru and app servers in Mumbai. Photos and PDFs sent in chats are still stored in New York.
  • PlannedAn uptime promiseOur uptime page shows measured figures. We will state a promise only after 90 days of measurements.

Every feature and its status is on the feature status page, and whether the service is working right now is on the uptime page.