here's the docs for getting started:
docs.firebender.com/context/conf...

**Better editing experience**. We added linting + improved our docs to be at parity with the plugin. This makes it much easier to modify the firebender.json and mess around

**Rules referencing markdown files!** JSON is limited for multiline strings.

Now you can reference existing ".md" files in your codebase!

more info here: docs.firebender.com/context/syntax

**Personal MCP servers**. MCP servers are growing in popularity - play around with them in Firebender

2 min walkthrough: docs.firebender.com/context/mcp

**Personal rules**. we all have different workflows, so now you can specify your own without changing the project firebender.json.

Tell the agent how you want it to work: "run custom detekt script terminal command after it making changes"

so far the benchmarks on others seem promising, but take with a grain of salt. we will release kotlin specific results to see how much real progress

openai.com/index/introd...

the benchmark take longer, we running them right now (much tokens and many container to make worky)

We also added claude 3.7 thinking agent mode as well! See all other features added and bug fixes at docs.firebender.com/about/change...

Based on our open source bench marks kotlin-bench, Gemini 2.5 Pro is the top performing model followed by Claude 3.7 Sonnet! firebender.com/blog/kotlin-...

just added gemini 2.5 pro support in agent mode too ;)

let me know if you run into any bugs or problems with the update - thanks!

Running build, test, and any other terminal command on its own without asking for approval each time

side note, I bought the Nothing Phone 3a on launch and its crazy good - display, camera, battery life, touch and feel. it's by far the best phone I've owned and also the most affordable one. super impressed w/
Nothing

Oh hell yes! Been looking for something like this for my KMP/CMP stuff! (and Jetbrains are dragging their feet with Junie)

First impressions after about an hour of using it.

1. Absolutely love how it fixes it's own errors. 😂
2. Autocomplete feels much faster than Copilot.
3. Very eager to make changes outside of the scope of the file I'm working on. Might be user error and might be fixable with rules

4. Sometimes it feels like it's doing things under the Hood, but I don't get an indication about it until the code in my open file changes. So not sure if it's working or hangs.
5. Multiplatform just works.