Hand Kalnavi a Goal: Agentic AI and Research That Shows Its Work
Most AI chatbots work one message at a time. You ask, it answers, and you stitch the pieces together yourself. That is fine for a quick question, but real work usually has a dozen steps between "I need this" and "it's done".
Kalnavi Agentic AI starts from the other end. You write the outcome you need, and a team of agents breaks it into steps, does the work, tests the result against your goal and fixes what falls short. Alongside it, Kalnavi's Research space applies the same idea to questions where sources matter.
How a mission runs
Every mission goes through the same four stages:
- Plan. Your goal is split into steps, and each step gets a benchmark it has to meet. You can review the plan before anything starts.
- Execute. Each agent builds on the previous agent's output, using web search and a document builder.
- Check. A reviewer scores the result against the goal you wrote.
- Fix. The biggest gaps are fixed automatically. Whatever is still missing is shown to you as it is, not hidden.
Here is what that might look like in practice. Imagine you run a small coffee roastery in Jakarta and type: "Should we start selling in Surabaya? I need a two-page brief on local competitors and their prices, ending with a recommendation." The plan would break this into steps such as mapping the roasteries already selling there before drafting the brief. Each step searches the web and passes its findings to the next. The reviewer then checks the brief against what you asked for, flags a missing section if there is one, and the gap is fixed before the file reaches you.
The example is ours, not a published case study, but the stages are exactly the ones Kalnavi describes for every mission.
Where the agents work
| Where | What happens |
|---|---|
| Missions | Market research, study plans, reports, slides, web pages and more, delivered as finished files |
| Playbooks | Tested starting points for common jobs. Replace the parts in brackets, then run |
| Scheduled missions | Agents work on their own every morning and send the result to your email and Telegram |
| Your browser | The Kalnavi side panel for Chrome reads the page, clicks and types to finish tasks in the tab, with you in control |
| Director on Telegram | On Enterprise, type /misi and your goal in Telegram. The agent team does the work and sends the result back to the chat |
| Email to Agent | On Usaha and Enterprise, forward an email to the agent and get the result as a reply |
The browser panel is the one most people try first. Install the free Kalnavi AI extension (our download guide shows how), open a page, and ask the panel to do something on it. It works in the same tab you are looking at, so you can watch every step.
You stay in control
Handing work to agents only makes sense if you can trust where they stop. Kalnavi draws two firm lines:
- Permission first. Any step that sends, buys, orders or deletes something waits for your approval.
- Your secrets stay yours. Kalnavi never types passwords or card numbers. Those remain in your hands.
Agents can still be wrong, and Kalnavi does not pretend otherwise. That is why every mission ends with its check results: what passed, and what still needs your attention.
Research with sources you can check
Research is where AI mistakes cost the most, so Kalnavi's Research space is built around showing where every claim comes from.
From question to findings
- Answers with citations. Answers drawn from the web list numbered sources, so you can open each one and check it.
- Long documents. Attach a PDF, a Word file or a dataset, then ask about what is inside.
- Research missions. Hand a topic to the agent team and receive a finished report with its sources.
Analyse and write
- Data Lab. Explore data with Python that runs right in your browser.
- Writing tools. Spelling checks and a web-similarity checker help you review drafts. The similarity checker compares against the web only and is not a Turnitin database.
- Knowledge Map. Keep what you learn in a map that belongs to your own account.
Kalnavi's own advice on reliability is worth repeating: treat AI answers as a starting point and check important claims against their sources yourself.
Kalnavi Code: a coding agent that proves its work
For developers, the same plan-and-check approach lives in Kalnavi Code. It works inside VS Code, reads your project and explains how it works, proposes edits you can review, then runs your tests and keeps fixing until they pass.
- Edit with approval. In manual mode, every edit and every command waits for you.
- Tests until green. It runs the tests, reads the failures and fixes them.
- Majelis review. Several models review your change before it ships.
- Your code stays local. Files and commands remain on your computer. The server only decides the next step.
- Safety for risky commands. Anything destructive is always asked first.
Setup instructions for every developer tool, VS Code included, are in our download guide.
Which plan you need
| Plan | Multi-agent tasks per day | Scheduled missions | Director on Telegram |
|---|---|---|---|
| Free | None | None | No |
| Pro | 30 | None | No |
| Bisnis | 100 | None | No |
| Usaha | 400 | 3 | No |
| Enterprise | 1,500 | 20 | Yes |
Multi-agent missions start on Pro, and scheduled missions come with Usaha and Enterprise. The Free plan is still a good way to meet Kalnavi: 20 chats a day with no card. Full prices are on the Pricing page and in our post on Kalnavi for Business.
To try your first mission, open kalnavi.com, choose Agentic AI, and write the result you want in one or two sentences. The more specific the goal, the sharper the plan.