Continue is an open-source AI assistant for VS Code and JetBrains. It reaches OpenAI-compatible servers through its openai provider with an apiBase override, as shown in Continue’s OpenAI provider docs. BroadNet Inference fits that pattern: same request shape, different base URL and model IDs.
Setup
- Get a BroadNet key from the console. See API keys.
- Open the local config. It lives at
~/.continue/config.yamlon macOS and Linux and%USERPROFILE%\.continue\config.yamlon Windows. In the IDE, open the configs dropdown above the chat input and click the gear icon next to Local Config. - Add the models.
name: My Config
version: 0.0.1
schema: v1
models:
- name: BroadNet GLM 5.3
provider: openai
model: broadnet/glm-5.3
apiBase: https://inference.broadnet.ai/v1
apiKey: your-broadnet-key
roles:
- chat
- edit
- apply
capabilities:
- tool_use
defaultCompletionOptions:
contextLength: 1048576
maxTokens: 8000
- name: BroadNet Qwen3.8 27B
provider: openai
model: broadnet/qwen3.8-27b
apiBase: https://inference.broadnet.ai/v1
apiKey: your-broadnet-key
capabilities:
- tool_use
- image_input
defaultCompletionOptions:
contextLength: 262144
maxTokens: 8000
- Save and pick the model in the model selector in the Continue panel.
Keep the file out of version control, because it holds the key.
What the fields do
provider: openaimakes Continue send standard chat-completions requests.apiBasereplaces the default OpenAI address. Don’t append/chat/completions; Continue adds the path.capabilitiesoverrides Continue’s autodetection. Its reference saystool_useis required for Agent mode andimage_inputenables image uploads. Continue can’t detect these for custom IDs, so set them.contextLengthandmaxTokenssit underdefaultCompletionOptions. Reasoning counts towardmaxTokenson BroadNet, so leave room.rolesdefaults to chat, edit, apply and summarize. We left outautocomplete: these models reason before answering, which suits chat and edits better than keystroke-level completion.
Which model to pick
| Use | Model | Why |
|---|---|---|
| Chat and Agent mode | broadnet/glm-5.3 |
Tools and structured output, 1M context |
| Screenshots, Arabic | broadnet/qwen3.8-27b |
Tools and image input, 262K context |
| Long chats on a flat plan | broadnet/deepseek-v4.1-flash |
Unlimited plans, 1M context; tools not listed |
For DeepSeek V4.1 Flash, drop tool_use unless your own test shows tool calls work, and set contextLength: 1048576. Prices for every model are on the pricing page. The GLM 5.3 model page has its limits.
Troubleshooting
- 401. Check
apiKey. A wrong, expired or out-of-allowance pay-as-you-go key returns 401. - Model errors on the first request.
modelmust be the exact BroadNet ID, with thebroadnet/prefix. GLM 5.3, Qwen3.8 27B and DeepSeek V4.1 Flash need a top-up; the $5 starting credit covers BroadNet Gemma 4 VL 26B and the two Transcribe models. - Empty or cut-off replies. Raise
maxTokens. Why answers come back empty. - 429. A pay-as-you-go key allows 30 requests a minute and 4 at once. Agent mode makes many calls in a row; add backoff in your workflow or space out runs. See errors and retries.
- Several agents at once on a plan key. Each unlimited plan session runs one request at a time. Give each parallel agent its own session.
- Wrong endpoint. Continue’s docs mention
useLegacyCompletionsEndpointanduseResponsesApi. Don’t set the legacy option; BroadNet serves/v1/chat/completions. Continue uses/responsesby default only for OpenAI’s o-series and GPT-5 models, so BroadNet IDs use chat completions without extra settings.
