An experimental open source project exploring a fully autonomous GPT-4 system.
Open source
24 alternativesADAS
ADAS explores ways to automate the design of agentic systems, as described in its research repository.
About ADAS
ADAS is a research code repository for Automated Design of Agentic Systems, a research area focused on automatically creating agentic system designs, including inventing new building blocks or combining them in new ways. It implements Meta Agent Search, in which a "meta" agent iteratively programs new agents in code based on previous discoveries.
The repository contains self-contained experiment folders for domains including ARC, DROP, MGSM, MMLU and math transfer, run through a search.py script per domain. Setup uses Python 3.11 with requirements.txt and an OpenAI API key, and the code can be adapted to new domains by editing evaluation functions and prompts. The repository warns that it executes untrusted model-generated code.
Key features
- Meta Agent Search where a meta agent writes new agents in code
- Self-contained experiment folders for ARC, DROP, MGSM, MMLU, math transfer
- Adaptable to new domains via evaluate_forward_fn() and prompt edits
- Python 3.11 setup with requirements.txt and OpenAI API key
- Apache-2.0 licensed; linked to an ICLR 2025 paper
Best for: Researchers and developers studying or extending automated design of agentic systems who are comfortable running Python experiments with their own OpenAI API key.
Good for
- Approach is documented in an ICLR 2025 paper and received an Outstanding Paper award at the NeurIPS 2024 Open-World Agent Workshop
- Experiment code is organized as self-contained per-domain folders
- Released under an Apache-2.0 license
Watch out for
- The repository executes untrusted model-generated code and warns such code may act destructively due to model limitations or alignment
- Requires conda, Python 3.11, dependency installation and an OpenAI API key to run
- No hosted interface; experiments are launched manually with search.py per domain
ADAS at a glance
- Website
- github.com/ShengranHu/ADAS
- Category
- AI & Machine Learning › AI Agents & Automation
- Type
- Open-source project
- Pricing
- Not confirmed. Check the site
- Open source
- Yes · Source code
- Language
- English
- Last checked
- Oct 6, 2026 · Online when checked
- Listed
- Oct 5, 2026
ADAS alternatives
Tools and sites you could use instead, ranked by how closely they match.
The Vision Agent repository is deprecated and directs users to Agentic Document Extraction.
Open source
24 alternatives
Agent4Rec implements research on generative agents for recommendation systems.
Open source
24 alternatives
SymbolicAI explores a neurosymbolic approach to large language models.
Open source
24 alternatives
BabyCommandAGI explores combining command-line interfaces with language model agents for task execution.
Open source
24 alternatives
AIlice provides an autonomous, general-purpose AI agent through its open-source repository.
Open source
24 alternativesPopular in AI Agents & Automation
Well-known listings from the same topic.
Simsim.aiMaya AI.smartmaya.aiMakeformsmakeforms.ioAgent Skillsagentskills.ioGenFuse AIgenfuseai.comBeambeam.ai
ADAS FAQ
What is Meta Agent Search?
It is the ADAS algorithm in this repository in which a "meta" agent iteratively programs interesting new agents in code based on previous discoveries.
How do you run an experiment?
Create the conda environment with Python 3.11, install requirements.txt, export your OpenAI API key, then run python {DOMAIN}/search.py for the domain folder such as _arc, _drop or _mgsm.
Can Meta Agent Search be adapted to a new domain?
Yes. Modify evaluate_forward_fn() and formatting prompts, consider adding basic functions for the meta agent, and update the domain-specific information in the prompts before running search and evaluation.
Is something wrong or out of date?
Tell us and an editor will check it.
Own this website?
Claim this listing
Claim the listing and suggest updates yourself. An editor checks every claim first.