Rig v0.34.0 now available!

[gold-silver-copper]

Hi everyone, Rig v0.34 is available to use!

Let's get right into talking about what's new this version and what's been updated.

Generic, immutable history

One of the bigger API improvements in this release is that prompt-generation methods such as prompt, stream_prompt, and related interfaces now accept iterators over &Message rather than owned or mutable Vec<Message> values.

If your application already owns and manages conversation state, you can now pass borrowed history into Rig directly instead of cloning, which should improve performance when working with large histories. It should also allow for easy streaming of messages from a database, without any intermediate cloning.

Better request customization for OpenAI-compatible providers

Rig now respects custom Authorization headers set via http_headers().

Previously, Rig would always overwrite the Authorization header with the provider's default auth scheme, which made some OpenAI-compatible backends awkward or impossible to use cleanly. In practice, that meant providers which expect something other than Bearer <token> often required workarounds outside of Rig.

With this change, if you explicitly provide an Authorization header yourself, Rig leaves it alone. This makes it much easier to work with custom gateways and OpenAI-compatible providers that use different auth conventions.

xAI TTS support

We've added Grok / xAI text-to-speech support in this release, further expanding Rig's model capability coverage.

Native TLS support

You can now choose native-tls instead of the default rustls, which should make integration a little easier for environments that prefer or require native TLS handling.

Telemetry and Anthropic caching improvements

Rig's OpenTelemetry GenAI cache token reporting is now more spec-compliant, which should improve compatibility with OTel-aware backends like Honeycomb, Grafana, and Langfuse.

This release also adds automatic Anthropic prompt caching helpers on the model builder, with support for both the default 5-minute TTL and an opt-in 1-hour variant where available.

Fixed

• Fixed an issue where Gemini enum schemas inside anyOf / oneOf would not infer string type correctly
• Fixed an issue where assistant text was missing from chat_history during multi-turn streaming
• Fixed an issue where encrypted_content was serialized even when it was None
• Fixed a VoyageAI usage deserialization failure when prompt_tokens was missing
• Fixed Gemini response parsing by making prompt_token_count optional

Other

• rig-derive now correctly propagates function visibility to generated tool structs

---

Full changelog: #1532

Thanks for reading! We build rig, rig builds us.

[joshua-mo-143]

Hi everyone! You are probably wondering why I haven't made the announcement post this time.

Today or so is actually my last day at Playgrounds, and as such I will be handing lead maintainership over to @gold-silver-copper who will be taking my place. He has a number of years of experience in open source, he's good at Rust and WASM and he also additionally has plenty of experience with AI so I foresee Rig being in good hands while he's in charge as i personally hand-picked him for the role.

(and yes, I realise it's April Fools Day... 😓 )

[gold-silver-copper]

Thanks Joshua, for building Rig into what it is today and all the work you've put into it, I hope to carry on your legacy as a top tier maintainer. You have truly been amazing, so it's big shoes to fill, but I will try my best!

Looking forward to working with everyone in the community and making Rig the best it can be. We have a lot of things we are working on behind the scenes, so stay tuned!