Warp is now open source

[zachlloyd]

Today we're announcing a change in how we build Warp: the client is now open source, and we invite the community to participate in building it using an agent-first workflow managed by Oz, our cloud agent orchestration platform. Full write-up on the blog: https://www.warp.dev/blog/warp-is-now-open-source.

The short version: we think the future of agentic development is wide open, and we want to figure out what the right workbench for building with agents looks like with the community rather than keep guessing at it privately. Oz does most of the implementation heavy lifting on approved issues. The community shapes direction and helps verify. The Warp team reviews and ships. You can take a look at the contribution flow in our CONTRIBUTING.md.

We want this discussion to be a home base for the big-picture conversation — what you think of the direction, what you'd want to build on top of Warp, what you're skeptical about. Bugs and feature requests belong in Issues, but for the meta stuff, this is the place. Thanks for being here — excited to build this together.

— Zach

[KaKi87]

So, no more login to open the terminal ?

[Pililink]

This feels like the right strategic move at the right time. With today's AI/vibe-coding capability, the advantage of a closed-source product with a similar UX can erode very quickly: people can now prototype, fork, and iterate on missing pieces much faster than before.

In that environment, the durable moat is less about keeping the client private and more about trust, review quality, product taste, extension surface, and the speed at which the community and maintainers can turn good ideas into shipped improvements.

Open-sourcing the client makes Warp much more credible as the workbench for agentic development. Excited to see how the contribution flow around Oz works in practice.

[ccraig]

This is fantastic news! Maybe now we can get local models to integrate!

[R3flector]

🚀 Congrats to the team and fair winds to Warp! It's a rare case when a product of this scale - a million active developers, five years of work - doesn't lock its work behind closed doors but instead entrusts it to the community.

For me personally, this isn't just a chance to "use a tool" - it's a chance to actually shape what a developer's workspace looks like in the age of agents. Truly grateful for the opportunity to take part.

To fellow community members: don't sit this one out. The future of agentic IDEs is literally being written in the open right now - and it will be exactly what we make of it. If everyone contributes even a small issue, a bug report, a patch, or an idea on the roadmap, we'll end up with a tool built around real-world needs, not someone's vision in a vacuum. Moments like this don't come often; it would be a shame to let it slip.

Good luck, Warp! 🚀

[kowyo]

Can we bring our own AI model and skip login?

[swein]

Would love to see open sourcing your full warp_proto_apis.

The Warp client does NOT directly call any LLM provider APIs. All AI requests flow through Warp's own backend at app.warp.dev via GraphQL. The client is an intelligent terminal with a beautiful UI, but the AI brain — LLM routing, model selection, billing, streaming orchestration, conversation state — lives entirely on Warp's servers. BYOK (Bring Your Own Key) keys are collected from users but forwarded to Warp's server, which then calls provider APIs on behalf of the user.

Pointing Warp at a custom OpenAI-compatible API (e.g., self-hosted vLLM) is difficult. The recommended approach is a Warp Backend Proxy that intercepts and translates GraphQL operations between the client and your endpoint — zero client modifications needed.

[marceldarvas]

Help docs already mention Fireworks as an AI provider.
I just want BYOK with Fireworks