Use Roo Code with BroadNet Inference (OpenAI Compatible)
Updated
Roo Code was shut down on 15 May 2026, so new setups should use Cline or the community fork Zoo Code. In existing installs, set API Provider to OpenAI Compatible, Base URL to https://inference.broadnet.ai/v1, API Key to your BroadNet key, and Model to broadnet/glm-5.3. Roo Code needs native tool calling, so pick a model with tools.
Roo Code was shut down on 15 May 2026. The repository is archived and read-only. For a maintained agent, use Cline with BroadNet Inference, or the community fork Zoo Code, which continues Roo Code’s codebase. The steps below still apply to existing Roo Code installs. See also Cline vs Roo Code vs Aider.
Roo Code is a coding agent for VS Code. Its OpenAI Compatible provider works with any endpoint that speaks the OpenAI chat-completions format, including BroadNet Inference at https://inference.broadnet.ai/v1. One rule shapes the setup: according to Roo Code’s docs, it uses native tool calling only, with no XML fallback. The model you pick must support OpenAI-style tool calls.
Open Roo Code settings with the settings icon in the Roo Code panel.
Set API Provider to OpenAI Compatible.
Enter the connection.
Field
Value
Base URL
https://inference.broadnet.ai/v1
API Key
your BroadNet key
Model
broadnet/glm-5.3
Adjust Model Configuration. Set Context Window to the model’s context from the table below. Set Max Output Tokens with room for reasoning; start with a few thousand. Turn on Image Support only for Qwen3.8 27B or DeepSeek V4.1 Flash. Enter Input Price and Output Price from the pricing page if you want Roo’s cost estimate to match your bill.
Run a test task, for example “list the files in this folder and summarize the README”. Roo Code will call its file tools. If those calls work, tool calling works.
Which model to pick
Model
ID
Context
Tools listed by BroadNet
GLM 5.3
broadnet/glm-5.3
1,048,576
Yes
Qwen3.8 27B
broadnet/qwen3.8-27b
262,144
Yes, plus image input
DeepSeek V4.1 Flash
broadnet/deepseek-v4.1-flash
1,048,576
Not listed; takes images
Start with GLM 5.3. It has a 1M-token context and BroadNet lists tools and structured output for it. Use Qwen3.8 27B when you want to paste screenshots or work in Arabic.
DeepSeek V4.1 Flash is the model on the unlimited plans. Because Roo Code needs native tool calls and BroadNet’s docs don’t list tools for it, test it before you rely on it. A Meteor hour costs $1 and is enough for a real task. BroadNet Gemma 4 VL 26B is also not listed with tools, so it is not a good fit for Roo Code.
Troubleshooting
401. The key is missing, invalid or expired, or the pay-as-you-go allowance is used up. Re-enter the key and check your balance.
The first request fails with a model error. The Model value must match a BroadNet ID exactly, with the broadnet/ prefix. GLM 5.3, Qwen3.8 27B and DeepSeek V4.1 Flash need a top-up; the $5 starting credit covers Gemma 4 VL 26B and the two Transcribe models.
Tool-calling errors. Switch to a model listed with tools. Roo’s docs note that partial tool support on the provider side causes these errors.
Empty answers. Reasoning counts toward the output limit. Raise Max Output Tokens. Why answers come back empty.
429. Pay-as-you-go keys allow 30 requests a minute and 4 at once. Roo’s subtasks and a second VS Code window share that limit. See errors and retries.
Parallel tasks on a plan. One plan session runs one request at a time. If you run an orchestrator and subtasks in parallel, each needs its own session, or they queue.
Questions
Is Roo Code still maintained?
No. Its GitHub repository was archived on 15 May 2026 and the README says the extension was shut down. It points users to Zoo Code (spelled ZooCode in the notice), a community fork, and to Cline. Existing installs keep working with the settings below, but they get no fixes.
Why does Roo Code show tool-calling errors?
Roo Code uses native tool calling only, with no XML fallback. Use a model BroadNet lists with tools, such as broadnet/glm-5.3 or broadnet/qwen3.8-27b.
Why does Roo Code get an empty reply?
The model used its output budget on reasoning. GLM 5.3, Qwen3.8 27B and DeepSeek V4.1 Flash reason first, and reasoning counts toward max_tokens. Raise Max Output Tokens in Model Configuration.
Can I use the unlimited DeepSeek plan with Roo Code?
The unlimited plans cover DeepSeek V4.1 Flash. BroadNet's docs don't list tool calling for it, and Roo Code requires tool calling, so test it on a Meteor hour before you buy a month.