FrogTeam.ai - Your Virtual Gen AI Dev Team

Update v0.2.7: 1-12-2025

lead-engineer tool error fix - the reason for this release
mlflow integration needs work but there are changes to start making this useful for tracking experiments

Update v0.2.5: 12-08-2024

Add save button for prompt and member panel (sometimes they get out of sync)
Team member names are now animals

Update v0.2.4: 12-08-2024

Properly queue requests (inside the extension) and throttle (goal: make configurable)
Finish making default configuration use anthropic.claude-3-5-haiku-20241022-v1:0 for team member roles

Update v0.2.3: 12-08-2024

Change to anthropic.claude-3-5-haiku-20241022-v1:0

Update v0.2.1: 12-07-2024

Focus on Anthropic
Full config out of the box for Bedrock hosted models defaulting to us-west-2
Use stability.stable-image-core-v1:0 for graphic artist
File locking fix
Suppress useless error messages

Update v0.1.6: 10-13-2024

Better event handling for saving prompts and setups
>Wrapper prompt for dall-e-3
Additional lead-architect prompt

Update v0.1.5: 10-01-2024

Add a new Team Member and make them a Graphic Artist
Choose the dall-e-3 model
This is a new feature there is no validation and no instruction

Prompt Example for using a Graphic Artist
I named my Graphic Artist Jenny. When I described what I wanted I also included the follow to explain Jenny's limitations and how I wanted the lead-architect to handle her assignments.

Make sure to ask Jenny to make graphics. Jenny's instructions must be very specific. You can only ask her to make one graphic file at a time and you can only describe what you want her to make in the prompt. A prompt for Jenny should be short, for example: "make me a small icon file that looks like a frog." Jenny just returns the path to the file she made. You need to work around her limitations. As the lead-architect plan out what you need from Jenny first and then tell the others what to do with what you had Jenny create!

Update v0.1.2: 09-08-2024

NEW TOOL: Code Search - If you want to make a change that can affect multiple files there is a new tool the LLM can use to search the code of the solution.
Better tool call error handling

Update v0.1.1: 09-08-2024

mlFLow experiments for tracking prompts
mlFLow config in .vscode/frogteam/config.json
moved frogteam files to .vscode/frogteam/
Fixed projects.jsonb file
Gave answer tab state

Update v0.1.0: 09-07-2024

mlFLow experiments early setup mlFlow only work from localhost:5001
Fixed webview post message events
updated member and prompt tree items

Updates v0.0.19: 09-01-2024

Bug Fix: task-summary type missing
Bug Fix: Duplicate and Delete buttons we erring
Updated included prompts

Updates v0.0.18: 08-30-2024

Project level at the top of the history tree
- Project tracked in history entry
Multi-Root Workspace Compatibility
Task a member directly from their configuration page
CSS and JS Webview consolidation
Proper content security policies

Updates v0.0.17: 08-25-2024

Commands - a top level menu entry
History hierarchy Changes
- Toggle History Grouping (See "Commands")
- Parent/Child elements but a flat tree
Respond to Answers directly
- In the History's Answer panel when the response is Markdown there is a "Respond Here" button
- When using this feature relevant immediate history will be included in the new LLM interaction
The Builder now collects a project name and directory
- This information is used to format XML that is used in the prompt
- This tells the LLM exactly what its getting
- System prompts will be adjusted in future versions
- The next version will use "Project Name" in the history hierarchy

Updates v0.0.16: 08-14-2024

Azure OpenAI
Upgrade Axios due to vulnerability report
Added some notes to the member setup panel

Updates v0.0.15: 08-13-2024

Tagging History Entries
Updated History Display

Next I am going to add a new tool allowing the LLM to query history as needed. I am also thinking about how to allow the LLM to query the user. If this query the user tool is used the conversation would have to suspend until the user replies. This is a feature that user may want to turn off. I'd love to hear some feedback about this.

Updates v0.0.14: 08-10-2024

GitHUb Repo Changes

Updates v0.0.13: 08-10-2024

Moving to GitHub Team: https://github.com/SpiderInk/frogteam
Lead Architect can use all implemented models.
Added Running Status Indicator in the StatusBar says "Frogteam" when its a project run and "member-name" when its a directed run.
Added an Output Channel called "FrogTeam.ai" that updates with every history entry and on other events.
Added New Member and Prompt commands to the project view to make these actions more visible.
Added Error message to tell yu when a team member has no aligned system prompt.
New prompt for requesting task/project summary
Wildcard prompts
import new prompts
API Key from Environment Variable

08-08-2024

Hi - Thanks for stopping by!

I decided to put this out there. Its at a nice spot where there is some functionality. The idea is to create team members that are represented by a specific LLM. You can use a variety of different LLM's and overtime how members are selected and how assignments are made will evolve from the rudimentary state I have it in today. You can use AWS Bedrock models and OpenAI models and for now I will likely stay inside these boundaries for LLM selection:

The model supports tool calling
The model and its tool calling feature are supported by langchain

I am currently focusing on some UI features while I enhance/refine my tool calling chain. I hope soon to move on to a system prompt sharing feature and eventually I would like to integrate RAG with local vectors. I hope then to turn around and use my extension to develop my next mobile app, whatever that may be.

I am wondering if there is an appetite for what I am doing here. Let me know your thoughts.

Here is a short demo video.

Your Virtual GenAI Development Team

This is meant to be a generative AI interface where you can register new models, configure model settings and prompts, interface with Chroma embeddings (eventually) and have a common set of commands you can use to interact with files in the workspace and the output from various LLMs. You register "team members" and assign them a model. You can use the same model multiple times or use different models. You can assign team members to collaborate on a task.

Overview

Lead Architect assignment
- The lead architect will break the project down and give other available members a task as it sees fit to the overall project.
- After all the assignments are done the lead architect will take a look in order to
  - fix any issues it can
  - summarize the overall solution
Single task assignments
- You can @member a task
Add/Remove/Edit Team member
- Name
- Model
- Limits (not implemented)
Prompt Library
- This will eventually link to a git repo or an https endpoint where new prompts will become available
  - This is planned as a community sharing platform
- Default Set
- Add/Remove/Edit
- tags: model, purpose

Planned Commands

@TeamMember

Flow

You must define at least three members:

lead-architect
lead-engineer
developer

Open the Builder

The "Builder" is found in the "Project" panel.

Project Description

Describe what the project is. Be specific and supply any information you may have. Code snippets, class hierarchy suggestions, research and example references. Web Site references for libraries you want to use, etc...

When you press "Project GO"

Lead Architect

The Lead Architect breaks down the project and

Provides each Member with an assignment

Each member works to completion on their assignment

When the task completes the information is added to the messages

When all members have performed their tasks the lead architect gets a final pass and summarizes the work that was done.

You can refine the prompt and submit again. Existing files will be used and edited.

Directed Assignment

In the Directed Assignment box you can call out a single team member to perform a task. This is nice if you like what has been done and now have follow up work that a member can do for you.

Known Issues/Limitations

At present this will only work directly with OpenAI or AWS Bedrock
- For OpenAI you need an API Key
- For AWS Bedrock you need to be logged into AWS in your VS Code Environment and you need the supported models deployed
Documentation is weak, I am working on it.
No tool call validation so sometimes initial project generation never completes you can try again
- validation/retry is coming
It is hard to tell when the work is done
- A status indicator is coming
Validation of Team Members to prompts, for now use Team Lineup view to manually validate that all members are aligned with a system prompt. If you see: TypeError: Cannot read properties of undefined (reading 'content') this is likely the issue.
If you paste into the prompt text area UI formatting may not work, it will save so just close and open the editor window for now.

Release Notes

0.0.11

lead-architect now supports Bedrock

0.0.10

Add region for AWS Bedrock

0.0.9

Marketplace categorization and copy updates

0.0.8

Icon change

0.0.7

Icon change
Copy updates in Prompt and Project views
Setting all default prompts to active

0.0.6 (and older)

This is an early version give it try. The following is what you can do.

Add Team Members and make sure they align with a prompt
- You can use openai with API Keys
- You can use Amazon Bedrock if your environment is logged in
You will get validation error messages if you don't have the minimum members/prompts
You can use the Team Lineup to validate Member/Prompt alignment
You can use the Builder to kick off the initial project creation
You can use the Builder to ask a specific team member perform a specific task
- You can say @membername can you edit the index.html file and change the title to "Narwhal"

Demo Video

How to Install

Open Visual Studio Code.
Go to the Extensions view by clicking on the Extensions icon in the Activity Bar on the side of the window.
Search for "FrogTeam.ai".
Click Install.
Or visit the Extension Store here.

Feedback

We would love to hear your feedback! Please provide any suggestions or issues you encounter on our GitHub Issues page.