GPTeam Collaborative Agents

A shared-world simulation where memory-equipped model agents communicate and work in parallel toward predefined goals.

Source repo
101dotxyz/GPTeam
Stars
★ 1.7k
Last updated
16d ago
License
MIT
Primary language
Python

At a glance

Works with
Portable with changesOpenAI API · Claude API
You'll need
PythonPoetryOpenAI API keyShell / CLINetwork accessLocal filesystem
Typical use
A developer studying multi-agent collaboration can observe memory-equipped agents divide work and communicate around predefined goals.
Main limitation
An OpenAI API key is required to run it, and optional tools require credentials for their respective services.

What does this agent do, and when should you use it?

GPTeam is an open-source multi-agent simulation centered on GPT-4. It runs multiple agents with individual memory inside a world where they communicate to pursue predefined directives. Agents can move between locations, perform tasks, speak with one another, and collaborate in parallel on common goals. While the world is running, text files in `agents/` summarize each agent’s current state. The project is launched from the command line and uses API keys in `.env` for model access and optional services.

The user runs python setup.py to check and configure the environment, adds an OpenAI API key to .env, and starts the simulation with poetry run world. The world runs multiple agents with memory that move among locations, perform tasks, communicate, and collaborate in parallel toward shared goals. During execution, agents/ contains one txt file per agent with a summary of its current state. To alter the simulation, the user changes agents or locations in config.json, runs poetry run db-reset, and starts the world again.

  1. A developer studying multi-agent collaboration can observe memory-equipped agents divide work and communicate around predefined goals.
  2. A technical team demonstrating GPT-4-based agent simulation can run the world and inspect per-agent state through the agents/ directory.
  3. An experimenter changing participants or locations can edit config.json, reset the database, and rerun the world with the revised setup.
  4. A cost-conscious user who accepts lower-quality outcomes can run poetry run world --turbo to use gpt3.5-turbo for every LLM call.
  5. A user with Anthropic credentials can set ANTHROPIC_API_KEY and use poetry run world --claude through the documented Claude path.

How do you install or deploy this agent?

  1. Clone the repository and enter the project directory: cd gpteam.
  2. Run python setup.py to check and configure the environment as needed.
  3. Add an OpenAI API key to .env; API keys for optional services enable other tools.
  4. Start it with Poetry: poetry run world.

How do you use this agent?

Run poetry run world to start and observe the agent world. For a cheaper run, use poetry run world --turbo; it uses gpt3.5-turbo for all LLM calls, with worse results expected. After changing agents or locations in config.json, run poetry run db-reset and then poetry run world. To use Claude, set ANTHROPIC_API_KEY and run poetry run world --claude.

What are this agent's strengths and limitations?

Pros
  • Models collaboration as a shared world with locations, movement, and agent-to-agent communication rather than a single model call.
  • Gives every agent memory and exposes per-agent state summaries in agents/ while the simulation runs.
  • Documents default GPT-4 execution plus --turbo, Claude, and Window execution options.
Limitations
  • An OpenAI API key is required to run it, and optional tools require credentials for their respective services.
  • The lower-cost --turbo mode is explicitly documented as producing worse results.
  • Window mode depends on the Window extension, and some models may respond slowly because prompts are long.
  • Changing the world configuration requires a database reset, which clears the existing database state.

How does this agent compare with similar options?

The default path uses GPT-4. poetry run world --turbo switches all LLM calls to gpt3.5-turbo for lower cost with expected quality loss, while --claude uses claude-v1 and claude-v1-instant for different calls.

Key facts side by side with the most closely related agents.

Agent Source review Stars Updated Language Full support on
GPTeam Collaborative Agents This agent 17 · Major gaps ★ 1.7k 16d ago Python OpenAI API · Claude API
Commonly 44 · Major gaps ★ 1.4k today TypeScript Codex · Claude Code
COMPASS Skills 85 · Good ★ 735 28d ago Python Codex · Claude Code
SwarmVault 67 · Some gaps ★ 694 2mo ago TypeScript Codex · Claude Code

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Major gaps
17/ 100 5-point scale 0.9 / 5
Trust 0/29
Reliability 3/14
Adaptability 4/18
Convention 4/18
Effectiveness 4/13
Verifiability 2/8
Why each dimension lost points
Trust0 / 29 · 0.0/5

Evidence shows: The repository provides no permission model, user confirmation mechanism, data flow transparency, sensitive data handling, dependency security audit, external effect control, rollback mechanism, or source attribution. All trust-related criteria are unsupported, hence score 0.

Reliability3 / 14 · 1.1/5

Evidence shows: There are some inconsistencies between README and configuration (e.g., README mentions `poetry run world`, but pyproject.toml defines `world` script, which is consistent; but README mentions `python setup.py`, while pyproject.toml does not mention it, and there is no setup.py file). Dependencies are listed in pyproject.toml, but no lock file or version pinning is provided, so dependency availability scores low. No error messages or failure handling documentation is provided, so failure messages score 0.

Adaptability4 / 18 · 1.1/5

Evidence shows: README describes target audience (researchers and developers exploring multi-agent collaboration) and scenarios (simulation). Capability boundaries are partially described (e.g., uses GPT-4, can switch models). Trigger precision is not explicit, as no clear trigger mechanism is defined. Environment fit is partially described (requires OpenAI API key, supports Claude and Window).

Convention4 / 18 · 1.1/5

Evidence shows: README provides project overview, installation steps, usage instructions, and contribution guidelines, but lacks FAQ, known limitations, changelog, and versioning. License is MIT, but no versioning or changelog is provided. Maintenance responsibility is not explicit, but contribution guidelines imply community maintenance.

Effectiveness4 / 13 · 1.5/5

Evidence shows: Output usability is partially described (agent states in agents/ folder), but no output format or quality assurance is provided. Marginal value is partially described (exploring multi-agent collaboration), but no comparison with existing solutions. Cost-benefit is partially described (using turbo to reduce cost), but no detailed analysis.

Verifiability2 / 8 · 1.3/5

Evidence shows: Claims in README are partially traceable to code (e.g., agents use memory and reflection), but no tests or independent verification are provided. Cross-source corroboration exists (blog post and video), but no detailed evidence. Fact and inference are not clearly separated, hence score 0.

Risks and how to mitigate them
  • Not found in source: least-privilege scopingGrant only what the task needs: a dedicated account or read-only token, scoped to specific directories and repos.
  • Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
  • Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
  • Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
  • Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
  • Not found in source: disclosed external effectsEstablish which external systems it writes to, sends to or changes, and verify with test accounts or repos before production.
  • Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
  • Not found in source: verifiable attributionInstall from the official repo or registry and check the publisher and URL to avoid look-alike packages.
  • The repository provides no security mechanisms such as permission control, user confirmation, or data flow transparency; use with caution.
  • Dependencies are not pinned and no lock file is provided, which may introduce supply chain risks.
  • Installation steps in README are inconsistent with pyproject.toml, which may affect reproducibility.
Evidence confidence: Low Reviewed Aug 09, 2026 Reviewed revision a216364ea2db
See the full review method →

FAQ

What credentials are required?
An OpenAI API key is required in .env. API keys for optional services enable their associated tools.
How can I reduce run cost?
Use poetry run world --turbo, which uses gpt3.5-turbo for all LLM calls; the documentation warns that results will be worse.
How do I see what agents are doing?
While the world runs, inspect agents/; each agent has a txt file summarizing its current state.
Can it run with Claude?
Yes. Set ANTHROPIC_API_KEY and run poetry run world --claude; the documented mode uses claude-v1 for some calls and claude-v1-instant for others.
What should I do after changing agents or locations?
Update config.json, run poetry run db-reset, then run the world again.
View on GitHub ↗ Install ↓

Related agents