CloudPilot

Inspiration

Deployments fail often and are time consuming to debug, we want to reduce the number of failed deployments for fast moving teams.

What it does

CloudPilot creates a map of your cloud environment and documentation, then lives in your IDE to help you write accurate config that adheres to your company's best practices, preventing the need to spend hours reading up on your cloud config's, talking to devops, or doing trial and error.

How we built it

• VSCode extension that can intake a developer request and passes the request and local directory to a server agent
• LangChain N-shot prompted model to understand and break down task to pass back to VSCode agent
• Using MindsDB API for inference

Challenges we ran into

• Interoperability (e.g. where to execute commands) between extension and agent not obvious

Accomplishments that we're proud of

• Building first VSCode extension
• Building an agent that can manipulate user files

What we learned

• Prompt testing to generate working config

What's next for CloudPilot

We're focussing next on...

• Improving the Knowledge Base: Ingesting business documentation, Jira/Slack interactions, and aggregating cloud resources to synthesize a knowledgebase we can pass in-context (or do RAG on top of)
• Building execution tools: Building the ability for Cloudpilot to add code, use the CLI, and even launch configs in Canary environments to do validation.

Built With

• agentops
• copilot
• langchain
• langserve
• mindsdb
• python
• typescript
• vscode