{% extends "base.html" %} {% block title %}{{ app_name }}: stop paying for agents that are stuck{% endblock %} {% block head %}{% endblock %} {% block body_class %}landing{% endblock %} {% block body %}
{{ app_name }}

AGENT SUPERVISION / PUBLIC BETA

More autonomy.
Less repetition.

Know when your agent stops making progress.
Get the evidence. Choose when to pause.

{% if allow_signup %}Start free {% endif %} See how it works
AGENT SUPRA / RUN EXPLORERILLUSTRATED DATA

RUN / GEMINI

Every attempt.
A clearer picture.

● No progress
ATTEMPTFAILING TESTSRESULT
011Baseline
024Regressed
031No new best
043No new best
FAILING TESTS 1 → 4 → 1 → 3 Four attempts. No new best result.
SUPERVISOR Paused The next step waits for your decision.
Illustrated run · Real Gemini test results
WORKS WITH YOUR STACK
GeminiOpenAIAnthropicPython SDK

Know when progress stops.

Track results across attempts. Start with observation, then enable pauses when you're ready.

  1. Install

    pip install agent-supra
    Works with Gemini, OpenAI and Anthropic.

  2. Connect

    Wrap your model client and report test results from your agent's loop.

  3. Stay informed

    Review evidence in your dashboard. Get alerts in Slack or email.

What it watches for

Readable rules. Choose which ones flag, pause, or stay off.

No progress

The same tests keep failing while the agent keeps changing the code.

Exact repeats

The same failure again with nothing changed, or the same tool failing the same way.

Going in circles

The agent flips between two versions of the code.

Context running away

Every call sends more tokens than the last, with nothing to show for it.

Budget and call caps

A hard limit per task, shared with any sub-agents it starts.

Fits your existing agent.

Use your own model client and API keys.

View a Gemini example
from google import genai
from agent_supra import (
    Client, SupervisedRun, SupervisorPaused)

sup = Client()   # reads your API key from env
task = sup.create_task("Fix checkout tests",
                       budget_usd=3)

with SupervisedRun(sup, task) as run:
    ai = run.wrap_gemini(genai.Client())
    try:
        while True:
            run.check()
            reply = ai.models.generate_content(
                model="gemini-3.5-flash-lite",
                contents=prompt)
            apply_fix(reply.text)
            t = run_tests()
            run.report_tests(t.all, t.failing,
                             rev=git_rev())
            if not t.failing:
                run.finish("success")
                break
    except SupervisorPaused as paused:
        print(paused.reason)
Read the quickstart

What it stores

Test results, token counts and fingerprints. Your prompts, code and model replies stay out. Privacy details.

Free while in beta

Agent Supra is free during beta. Model usage is billed separately by your provider.

A few good questions.

Will it stop my agent automatically?

Start in observation mode to record signals while your agent continues. Enable pause rules when you're ready; the SDK then pauses the next supervised step.

Do I need to change model providers?

No. Keep your Gemini, OpenAI or Anthropic client and your own API keys. Add Agent Supra to your existing loop.

Does it store my prompts or code?

It records test results, token counts and fingerprints. Prompts, code and model replies are not stored. Read the privacy details.

What does it cost?

Agent Supra is free during beta. You pay your model provider separately for model usage.

Give your agents
a little supervision.

{% if allow_signup %}Start free {% endif %}Read the docs
{% endblock %}