AutoGPT DRAFT : Project Abstraction Layer

UNFINISHED do not accept, this is to share advancements

Background

Implementation of ideas regrouped in #3392 for the project level. The implementation should be fully backward compatible. The implementation migrate the current ai_settings.yaml to a project structure but do not change the core , it change the back end.

Now = AutoGPT can have 1 agent
New back-end = AutoGPT can have 1 project with 1 agent
Future = AutoGPT can have N project with 1 agent each
Future ++ = AutoGPT can have N project with N agent each

The new back-end is further detailed in REAME.md

AIConfig, is an empty class that inherit from AgentModel (for backward compatibility without changing core variable name)
AgentModel, is a representation of an Agent and manage CRU of an Agent.
Project, is a common goal to various Agent and manage CRU of this project and D of Agent (not yet implemented)
ProjectsBroker, create & manage projects

Changes

➕Documentation : Created a projects/readme.md to help user & developper alike to work on the projects
➕Tests : 3 new pytest files
⛔️ AIConfig : Fully removed and kept for backward compatibility so the core is not modified
➕ AgentModel : Added
➕ Project : Added
➕ ProjectsBroker : Added

Changes that affect core :

⚡ prompt.py a consequence of the abstraction layer was that def construct_full_prompt(ai_config, prompt_generator: Optional[PromptGenerator] = None) -> str: had to be moved out AIConfig
⚡ 1 line in main.py to reflect the change above

Risk :

Manage circular imports as the backward compatibility requires imports from Agent to ProjectBroker & the setup is designed to work with imports from ProjectBroker to Agent.

Documentation

Every method will be fully doc stringed
README.md created
Variable created in .env template should be documented if any
Any twitch inserted for backward compatibility must have a # TODO and if possible PROJECT-DEPRECATED tag.

Test Plan

Unit test with pytest
Publication of the branch on Discord & request for testers.

Scenarii :

MVP :

[x] Start a first instance of AutoGPT without ai_settings.yaml
[ ] Start an instance of AutoGPT with ai_settings.yaml
[x] For any of the instance above save & quit after after at least 1 interaction with the agent
[ ] Start the instance above and see if the agent can load (type y) & quit after after at least 1 interaction with the agent
[ ] Start the instance above and see if we can create a new agent(type n) & quit after after at least 1 interaction with the agent

PR Quality Checklist

[x] My pull request is atomic and focuses on a single change.
[ ] I have thoroughly tested my changes with multiple different prompts.
[x] I have considered potential risks and mitigations for my changes. FULL BACKWARD COMPATIBILITY
[x] I have documented my changes clearly and comprehensively.
[x] I have not snuck in any "extra" small tweaks changes

Additional informations :

To loop back on features gathered in #3392 :

[x] AgentTeam object
[x] Dynamic / Extensible Agent (the agent is constructed dynamically allowing flexibility. & plug-ins)
[ ] Dynamic / Extensible Project
[ ] Dynamic / Extensible Agent Team
[ ] Status
[ ] State awareness
[ ] Implement folder per agent
[x] Agent unique ID
[ ] Agent budget
[ ] Agent budget consumption
[ ] Project open_api_key

YAML Option implemented : Option 3

my_project/
├─── team
│          ├── leader : 
│          ├─────────────  agent 1
│          └── specialized_agents : 
│          ├─────────────   agent 2
│          ├─────────────   agent 3
│          ├─────────────  agent 4

Folder Option implemented : Option 1 , could be a MVP

autogpt
├── projects
│      ├── project_1
│      │   ├── settings.yaml
│      ├── project_2
│      │   ├── settings.yaml
│      └── ...

Apr 29 '23 22:04 ph-ausseil

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
docs	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	May 6, 2023 10:39pm