DataBassGit/AgentForge

AgentForge – Low‑code framework for building AI‑powered autonomous agents

What it is – AgentForge is an open‑source Python library (≥ 3.12) that lets you describe agents, their memory, and the way they interact in cogs (YAML‑based workflow files). It is deliberately “low‑code”: most of the wiring is done declaratively, while you can still drop in custom Python code when needed.

Core ideas

• Agents – a single LLM‑backed chatbot defined by a prompt template and a configuration file.
• Cogs – YAML documents that orchestrate one or more agents, add branching logic, and attach memory nodes. Think of a cog as a reusable workflow diagram.
• Memory – optional persistent context that agents can read/write during a session, automatically made available to the agents referenced in a cog.
• Personas – separate YAML files that set an agent’s identity, tone, and background.
• LLM‑agnostic – each agent can point to a different model provider (OpenAI, Google Gemini, Anthropic Claude, or any local model served through Ollama/LMStudio).

Key features (as listed in the README)

• Declarative Cogs for multi‑agent orchestration and branching logic.
• Customizable agents via prompt‑template YAML.
• Integrated, optional memory that persists across calls.
• Persona files to control style and context.
• Dynamic prompt editing at runtime – no need to restart the process.
• Support for major LLM APIs (OpenAI, Google, Anthropic) and local models via Ollama or LMStudio.

Typical workflow

1. Install the package from PyPI (pip install agentforge).
2. Run the quickstart guide, which scaffolds a hidden .agentforge/ directory.
3. Write a simple cog YAML that references one or more agents and (optionally) a memory node.
4. Execute the cog with the CLI; AgentForge contacts the configured LLMs, feeds prompts, stores/retrieves memory, and returns the final output.
5. For production use, replace the demo “debug” model with a real API key (OpenAI, Gemini, Claude, or a local Ollama model).

Who might use it

• Researchers prototyping multi‑agent cognitive architectures.
• Developers building chat‑based assistants that need persistent context.
• Teams that want to experiment with different LLM providers side‑by‑side without rewriting code.
• Educators demonstrating agent orchestration concepts with minimal boilerplate.

Getting started

• Docs hub: docs/README.md – the central navigation page.
• Quickstart: docs/guides/quickstart.md – install, scaffold, run a no‑credential smoke test.
• First real model run: docs/guides/first_real_model_run.md – configure an API key or local model and run a real LLM.
• Advanced reference: docs/guides/advanced_reference.md – deep dive into agents, cogs, memory, storage, utilities, and legacy tools.

Community & contribution

• Issues and pull requests are welcomed.
• Contact via email (contact@agentforge.net) or Discord (link in README).

License – GNU GPL‑v3.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.