OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Agent Frameworks

Self Improving Agents

A Python research repository comparing four progressively advanced approaches to self-improving code agents: benchmark optimization, reviewer feedback, self-modifying source code and adversarial agent co-evolution.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Self Improving Agents — OpenViST project profile in Agent Frameworks.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when you want a concrete research reference for comparing self-improvement mechanisms rather than a single opaque agent product.

Who it is for

  • Researchers exploring self-improving code agents
  • Developers studying benchmark-driven agent loops
  • Readers comparing feedback, self-modification and adversarial evaluation patterns

Documented uses

  • Run documented agent levels against example coding tasks
  • Compare improvement strategies using logged experiments and analysis
  • Study how reviewer feedback, code rewriting and adversarial tests change an agent loop
  • Use the repository as a starting point for benchmark-driven code-agent experiments
Before you choose

Requirements & limits

The repository is public, active in the supplied record and MIT-licensed. It is presented as a CLI-oriented Python project and documents a Gemini API key requirement for the setup.

  • The four levels represent materially different architectures, from a single benchmark loop to adversarial co-evolution.
  • The documented examples cover Snake game AI, email validation and customer support.
  • Feedback Loop is described as a cost-performance trade-off, while Arena Loop is intended to expose brittleness through adversarial testing.
  • The repository documents checkpoint and resume support for interrupted experiments.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up and run a documented agent-level experiment

Install the repository dependencies, configure the documented API key, and run one of the available agent levels against a supported task.

Before you start

  • Python environment for the repository
  • The repository cloned or otherwise available locally
  • A Gemini API key placed in a root .env file as GEMINI_API_KEY

Steps

  1. Install the dependencies with pip install -r requirements.txt.

  2. Create a .env file in the repository root and add GEMINI_API_KEY=your-api-key-from-https://aistudio.google.com/apikey.

  3. Run a level with its documented command, such as python autoresearch/run.py or python feedback-loop/run.py.

  4. Select a documented task with the --task option, such as snake, support or email_validation where supported.

  5. For logged experiments, use the documented experiment.py or run_all.py commands and then analyze_results.py for cross-level comparison.

Expected outcome

A selected agent level runs against a documented task, with experiment logging and comparison available through the repository's analysis commands.

Guide limitations

  • The evidence documents commands and supported examples but does not establish results for every level-task combination.
  • The repository description does not provide implementation details for adapting the loops to an arbitrary benchmark.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
30 September 2026
Original website title
BetterForAll/self-improving-agents · Official homepage title · retained in its source language
How we review sourcesSuggest a correction