Why LimaCharlie’s AI Sessions works with any model

I have been using AI coding tools since the beginning. Back around 2022, I built a RAG system that would return links to relevant documentation when users made a search request. Initially, I wanted the AI to answer the user's question directly, but at the time it would hallucinate so much that I didn't trust the output enough to put it in front of users. Instead, I had the AI return static links to the relevant documentation.

It worked pretty well, but it was also a bit of a dead end in terms of something we would continue investing in.

Years later, Claude Code was the tool that made me realize something had fundamentally changed. These were the early days of what would eventually become known as vibe coding. Without any real knowledge of particular Python libraries, I could quickly start using them, deploy what I needed, and get the results I wanted.

Early on, Anthropic made a very intentional push toward building tools for developers. The difference between Claude and the other models in coding was dramatic, and that is why Claude was the first model we chose to integrate natively.

Over time, the other frontier labs started to see the same opportunity. Not only were these models extremely effective in the coding space, but developers were also willing to pay for API access and increasingly expensive monthly subscriptions to use them.

More recently, I have started getting reports from some very talented and well-known hands-on-keyboard practitioners that OpenAI's Astra is outperforming Claude for some of their work. To me, that is a good indication that this is not a race any one frontier lab is going to win outright. There will be a constant ebb and flow between the leading models as these companies continue to push each other forward.

To that end, it makes me incredibly happy that LimaCharlie AI Sessions are not tied to a single model or even a single way of accessing that model. For OpenAI, you can connect directly with an API key or use your Azure OpenAI deployment. For Gemini, you can use an AI Studio API key or your existing Google Cloud account through Vertex AI. For Claude, you can connect a Claude Pro or Max subscription, use an API key, or run through AWS Bedrock or Google Vertex AI.

We also support OpenRouter. With a single OpenRouter API key, you can access a much wider variety of models across different vendors, with the usage billed to your OpenRouter account.

This is important because I don't think there is going to be one model that permanently wins. The best model for a particular job is going to change, and people are going to have different preferences for how they access and pay for these services. Some already have Claude subscriptions. Some are heavily invested in AWS, Azure, or GCP. Others want to move between models through something like OpenRouter.

It has always been possible to use essentially any AI model to operate on the LimaCharlie platform through the CLI or MCP server. What AI Sessions changes is that you can now bring these models directly into the LimaCharlie web application as interactive operators working alongside you.

Rather than picking a winner for you, we want to make sure that as the frontier moves, you can bring the models and infrastructure you already use into LimaCharlie and put them to work.