Build and tune your own assistants, safely
Create your own assistants and adjust built-in ones to your business, within safety limits that can't be removed.
Live in partWhat does BYOA + Agent Studio do? AI shaped to your business without custom builds or security holes, with every change to a built-in agent recorded and approved by a person.
How it works
Admins build agents from tools Kinnow has checked. Safety checks run when you save, limiting what data an agent can reach and which AI models it may use. Built-in agents can be adjusted, but safety settings stay locked. Every change to a built-in agent is proposed with a reason and becomes the approved setting only after a person approves it. Agents you build yourself are ready to use once saved. Connecting approved Agent Studio settings to the running agent is rolling out with early customers.
A change that touches a locked safety setting is refused by name, and an unapproved change is never treated as the approved setting.
Step by step
Say the finance head wants an assistant that answers questions from project documents, and also wants to adjust how a built-in agent behaves. These are two separate screens.
- Build an assistant. An administrator opens Build agents, names the assistant and states what it knows, for example Drafts quotes from project documents.
- Give it instructions and limits. Instructions are written in plain language. Then choose What it can read from a list of tools that Kinnow has already checked, add rules one per line, and decide whether its answers are reviewed before they go out.
- Safety checks run when you save. Kinnow limits what data the assistant can reach and which AI models it may use. Reaching outside Kinnow needs a senior administrator and can never send anything on its own.
- It shows up in Ask Kinnow. The assistant appears in the same list of helpers as the built-in ones, with a version number that goes up each time you edit it.
- Tune a built-in agent. In Agent Studio, enter the agent's id, choose a different model, prompt version or setting, write why, and press Propose change. It sits as Awaiting approval.
- A person approves. Approve makes it the approved setting for your company only. Reject throws it away. A company can require that the approver is not the person who proposed it.
See an assistant built for your own paperwork. Tell us the two questions you ask most often.
What you see on screen
Both screens are in Admin.
- Build agents: a list of your assistants with Enable, Disable and Delete, and a form with What it knows, Instructions, What it can read, rules, Review before it goes out, and When it runs (on request or on a schedule).
- Agent Studio: the agent's current settings, the number of your own test cases kept as a benchmark, the proposed change and its reason, and the Active and Awaiting approval states side by side.
- Locked list: a section headed Locked (cannot be changed here) naming the safety settings that are the same for every business.
- Refusal by name: if a proposal touches a locked setting, it is refused and the setting is named. It is not quietly dropped.
Who uses it
| Who | When | What they do |
|---|---|---|
| Administrator | At set-up, and when needs change | Builds assistants, proposes changes, switches assistants on and off. |
| Senior administrator | When an assistant needs to reach outside Kinnow | Grants that access, which is never automatic. |
| Finance head or owner | Before a change is approved | Approves or rejects tuning changes. |
Related
- Admin: Build agents, Agent Studio and the other company settings.
- Agent guardrails: the limits every agent runs under.
- Explainable agents with human sign-off: how agent work is shown for approval.
- AI confidence that means something: how approvals sharpen confidence.
Questions
What is BYOA + Agent Studio in Kinnow?
Create your own assistants and adjust built-in ones to your business, within safety limits that can't be removed.
What changes for the business?
AI shaped to your business without custom builds or security holes, with every change to a built-in agent recorded and approved by a person.
Is it available today?
It is live in part: the core is built and parts are completing with early customers.
Can I make an agent send emails on its own?
No. Reaching outside Kinnow needs senior administrator rights, and an assistant that reaches outside can never send on its own. The review-first setting keeps a person in the loop.
Who can change a built-in agent?
Someone with settings-write access can propose a change, and the same permission approves it. A company can require that a second person approves. Safety-critical settings cannot be changed from that screen by anyone.
More in the AI agents that do the work
Claim every rupee of GST input credit
Purchase bills are matched against your GSTR-2B file from the GST portal; your team only handles the exceptions.
AI agents that do the work · Collections AgentOverdue invoices chased for you, politely
Overdue customers get reminders that grow firmer as days pass, each one drafted for a person to approve before it goes out.
AI agents that do the work · GatherForms for applications, surveys and sign-offs
Collect applications, data, anonymous feedback or sign-offs, each with its own AI helper and privacy rules.
AI agents that do the work · Close Checklist agent + 12-gate period closeMonth-end close that runs itself
Kinnow runs the month-end checks, plans the fixes for what fails and hands you one close pack. A person closes the month.
AI agents that do the work · Agent runsAI work you can check and approve
Each agent task has a Show me why view and shows its confidence where the agent recorded one; you approve it, reject it, or note that it looks wrong.
Tell Kinnow what slows you down.
Answer three questions. Get the applications, the agents and a two-week plan for your business.