Skip to content

AdityaModel

Your project. Real tools. Your decision.

A focused workspace being developed for exploring a codebase, following an investigation, and reviewing the work before a consequential action.

Aditya is being tested by our internal team. Email Aditya Lab about access. The interfaces below are illustrations, not a live model session.

What is Aditya, and can I use it now?

Aditya AI is an Indian-language AI research program. Aditya is being tested by our internal team; email the lab to discuss access.

Is there a general-access date?
No general-access date has been announced. Contact Aditya Lab about a specific evaluation.
Are these live responses?
No. The website shows interface illustrations and example conversations. They are not a live session or a benchmark.
What can I inspect today?
Read the evaluation, runtime architecture and source-build documentation. Those materials have their own scope and limitations.

Built to enter
the work.

Software work rarely stays inside one prompt. A failing test may lead through configuration, a call site, and a decision made months ago. Understanding the problem means following those connections, not just producing a plausible answer.

Aditya Code is a workspace around the agent runtime. You start with a project and a question. The runtime can read files, search the codebase, and use the tools available in that environment; the workspace brings the resulting activity and proposed changes into view.

For an investigation, that might mean finding where a value is set, following it to the failing code, and checking the explanation against a test. For a change, the useful result is a reviewable diff with the checks that ran—and any checks that could not be run.

The model can still miss context or suggest the wrong fix. Tool results are evidence to examine, not a guarantee of correctness. Approval rules depend on your configuration, so review the proposed action and its scope before letting it affect your project.

One intelligence.
Wherever work happens.

Interface illustrations. The project names, results, and run events below are examples—not a live session or model evaluation.

Keep the investigation in view.

The desktop workspace gives an investigation room to breathe: task history, source activity, proposed edits, and the checks beside them.

Request this surface
Project / atlas— □ ×
EXAMPLE INVESTIGATION

One possible cause.

The session guard may be skipped after client-side navigation. The next step is to check the route and its tests.

Read relevant filesFILE ACTIVITY
– return navigate('/login')+ return session.recover().then(retry)
Illustrative taskaditya / workspace

Trace the authentication failure, patch the cause, and show me the diff.

example prompt

Continue from a browser.

The web workspace presents the same run through a browser. What you can do still depends on the project, configured tools, and permissions.

Request this surface
aditya.local / project / atlas
dityaMODELWORKSPACE PREVIEW

EXAMPLE WORKSPACE

The work follows you.

A browser view of the project task, with its history and source context within reach.

Inspected repositoryRelevant files located
Diagnosed session raceCause under review
Preparing a proposed patchAwaiting review
Example reviewProtected actions require review

Start where the code lives.

The current CLI starts in a chosen project directory. It is useful when you want a compact, keyboard-first view of the run and its tool activity.

Request this surface
CURRENT CLI / EXAMPLEatlas / main
python3 -m agent_system.cli --workspace ./project

ILLUSTRATIVE RUN

01Read the workspaceRelevant files selected
02Trace the failurerouter/session.ts
03Check the proposalTest output to review
PROPOSED CHANGE

A cause to check.

The route and session code suggest a recovery race. The proposed change still needs review against the real project.

Diff to reviewChecks to inspect
Protected actions can pause for approval.EXAMPLE OUTPUT

Begin with the project, not a blank chat.

A coding task has a location. It may be a failing test, a route that behaves differently in production, or a change that touches several files. Aditya Code starts from a selected project so the investigation can refer to actual files and tool results.

As the run progresses, the useful question is not simply whether the agent has an answer. It is whether you can follow how it got there. Which files were inspected? Did a command complete? What did the test actually say? A failed check should remain a failed check, not become a confident sentence about success.

That trail also makes review easier. You can challenge an assumption, inspect a proposed diff, or ask for another check without having to reconstruct the whole journey from the final message.

When evidence is missing

If a check cannot run because a dependency or service is unavailable, the result should stay unverified. A plausible patch is not the same as a tested one.

An answer is not
the whole story.

ADITYA / WORKSPACE EXAMPLE WORK TRACE

PROJECT CONTEXT

Begin with the source.

A useful investigation starts by locating the code path, the nearby tests, and the conventions already used in the project.

01Search the project
02Read the relevant files
03Check existing patterns
CONTEXTACTIVITYPOLICY

Some actions need
a human answer.

Reading a file is different from deleting one or sending data outside the project. Aditya Code treats those as different kinds of authority.

AN EXAMPLE HANDOFF

A protected action is waiting.

The request should show what will happen and where. The operator makes the decision; the run records the outcome.

DECISION REQUIRED
ClassifyThe runtime checks the tool and the scope of the action.
PauseA protected action waits instead of silently continuing.
DecideThe operator can approve or deny the requested action.
RecordThe decision becomes part of the run history.

What exists now, and what is still being built.

Aditya Code brings together an existing local agent runtime and several ways to use it: a Python CLI, a terminal interface, a web workspace, and a desktop app. The runtime has project tools, run records, approval policy, and support for configured MCP connectors.

In the current system

Tool calls and approval decisions can be recorded in the run. The workspace can show activity and proposed changes, and the operator can review checks rather than taking a completion message as proof.

Still in progress

Aditya branding is being introduced across the product. The Python CLI uses python3 -m agent_system.cli --workspace. Other terminal packages can use different entry points; do not substitute an undocumented alias. Neural consolidation in NIF remains research-stage.

This is pre-release software, not a claim of frontier-model performance or unattended production readiness. The right expectation is a system you can inspect and evaluate in your own project, with its limits visible.

Where the workspace helps.

These are kinds of engineering work the system is built to support, not promises that every task will succeed without review.

Understand an unfamiliar codebase

Find the entry points and conventions before deciding what to change.

Trace a production failure

Follow the evidence through logs, source, and a reproducible check.

Change more than one layer

Keep related edits and their tests together for review.

Work with connected tools

Use configured MCP connectors when a task needs context outside the repository.

Keep documentation aligned

Update the explanation when an implementation changes.

Frequently asked questions

What is Aditya AI?

Aditya AI is an Indian-language AI research model and assistant from Aditya Lab. The website's chat and workspace examples are illustrations, not live model conversations.

Which languages does Aditya AI support?

Our evaluation covers Hindi, Gujarati, Marathi, Bengali, Tamil, Telugu and English. That measured scope does not guarantee the same quality for every request or language.

Does it understand romanised Hindi or Gujarati, including Hinglish and Gujlish?

The Indic evaluation includes Hindi and Gujarati written in Latin letters as well as native script. These saved tests do not establish that every Hinglish or Gujlish phrasing will work reliably.

How can I get access to Aditya AI?

Aditya is being tested by our internal team. Email [email protected] to discuss your use case and access; no general-access date has been announced. API-key calls are paid per token from a prepaid balance.

How is Aditya AI different from ChatGPT or Sarvam?

Aditya Lab focuses on Indian-language behavior and visible, reviewable agent work. Our published comparisons use sarvam-m as a base model on named evaluation suites, not as a general ranking against Sarvam or ChatGPT.

What do the research pages show?

The research pages distinguish saved measurements from architecture ideas and illustrative interfaces. Reports include methods and limitations so readers can interpret the results.

Is my data used for training?

Your conversations are not used to train or improve our models unless you opt in through your account settings; this is off by default. You can turn it off at any time, after which new conversations are not used. See the privacy policy for retention and connected-service details, and do not email sensitive material.