Most AI coding platforms work like this: your code goes to a server, a model processes it, results come back. You're trusting a third party with your source code, your API keys, and your development workflow. For hobby projects, that's fine. For real work, it's a problem.
In crewswarm, all execution happens on your machine. When an agent writes a file, it writes to your local disk. When it runs a command, it runs in your shell. When it calls an LLM, it goes directly from your machine to the provider's API — no proxy server in between.
This is a design choice with concrete consequences:
~/.crewswarm/crewswarm.json with file permissions you control. Not in someone else's database.When you're not locked into a platform, you can choose providers based on merit. crewswarm supports 24 providers out of the box — from Anthropic and OpenAI to Groq, DeepSeek, Cerebras, Together, and local inference via Ollama, vLLM, or SGLang.
More importantly, you can mix providers per agent. Your PM uses Groq for speed. Your coder uses Claude for quality. Your researcher uses Perplexity for web search. Switch any of them with one config change, no migration needed.
When a new model drops — and they drop every week — you update a string in your config. No waiting for your platform vendor to "add support." No feature flags. No upgrade tier.
Cloud AI platforms ask for broad trust: send us your code, your prompts, your file system access. You trust that they don't log it, don't train on it, don't get breached.
Local-first inverts this. You trust the LLM provider with your prompts (unavoidable — the model needs to see the code to work on it). But you trust nobody else with your execution environment, your file system, or your infrastructure.
The orchestration layer — the part that decides what code to write, which files to modify, what commands to run — runs on your machine, under your control, with source code you can audit.
Local-first isn't always the right answer. If you want a hosted product you never think about, a managed team deployment with centralized billing, or a mobile-first experience — cloud platforms serve that well.
crewswarm is for developers who want control. Control over which models they use, which providers they pay, where their code runs, and how their agents behave. If that's you, local-first is the only architecture that makes sense.