Vora Compute

Set up your agent

Use Vora models in your coding agent or chat app. Choose your tool, copy the settings and try a small task.

1. Get your Vora key

Open API keys, create a key and copy it. It is shown once. No account yet? Subscribe for Chat and API access.

For a desktop app, paste the key into its API key field. For a terminal tool, replace vora_sk_... below with your key and run this in the same terminal you will use to launch the tool:

export VORA_API_KEY=vora_sk_...

This sets the key for that terminal session. Keep it out of project configuration and source control. Continue uses its own secrets file, described below.

2. Choose a model

Use OpenAI Compatible / Chat Completions as the provider type and this base URL, including /v1:

https://api.voracompute.com/v1

Start with Qwen3.8 27B if you are unsure. The examples register deployed chat models where the tool supports it. Pending models are excluded until acceptance is complete.

Model IDContext tokensMax output tokensReasoning effort
qwen3.8-27b1,000,00032,768none / low / medium / high
nemotron-3.5-super-vl-120b1,000,00032,768none / medium / high
eurollm-22b32,76816,384none

Context includes your instructions, tool definitions, files, conversation history and output. Maximum output includes reasoning and the final answer. Use these Vora deployment limits when your tool asks for context or output size. Keep tasks focused and compact or start a new conversation before the context fills up.

OpenCode

  1. Set VORA_API_KEY in your terminal using the command above.
  2. Save this as opencode.json in your project, or merge the vora entry under providers into your existing configuration. For all projects, use ~/.config/opencode/opencode.json. This example uses the current OpenCode configuration format.
  3. Start opencode from that terminal. Run /models and choose a model under Vora Compute.
{
  "$schema": "https://opencode.ai/config.json",
  "providers": {
    "vora": {
      "package": "@opencode/ai/providers/openai-compatible",
      "name": "Vora Compute",
      "env": [
        "VORA_API_KEY"
      ],
      "settings": {
        "baseURL": "https://api.voracompute.com/v1",
        "headerTimeout": 3660000,
        "chunkTimeout": 3660000,
        "timeout": 3660000
      },
      "models": {
        "qwen3.8-27b": {
          "name": "Qwen3.8 27B",
          "limit": {
            "context": 1000000,
            "output": 32768
          },
          "capabilities": {
            "tools": true,
            "input": [
              "text"
            ],
            "output": [
              "text"
            ],
            "reasoning": {
              "supported": true,
              "toggle": true,
              "effort": [
                "none",
                "low",
                "medium",
                "high"
              ]
            }
          },
          "compatibility": {
            "reasoningField": "reasoning_content",
            "maxTokensField": "max_tokens",
            "streamUsage": true
          }
        },
        "nemotron-3.5-super-vl-120b": {
          "name": "Nemotron 3.5 Super VL",
          "limit": {
            "context": 1000000,
            "output": 32768
          },
          "capabilities": {
            "tools": true,
            "input": [
              "text"
            ],
            "output": [
              "text"
            ],
            "reasoning": {
              "supported": true,
              "toggle": true,
              "effort": [
                "none",
                "medium",
                "high"
              ]
            }
          },
          "compatibility": {
            "reasoningField": "reasoning_content",
            "maxTokensField": "max_tokens",
            "streamUsage": true
          }
        },
        "eurollm-22b": {
          "name": "EuroLLM 22B",
          "limit": {
            "context": 32768,
            "output": 16384
          },
          "capabilities": {
            "tools": false,
            "input": [
              "text"
            ],
            "output": [
              "text"
            ],
            "reasoning": {
              "supported": false,
              "toggle": true,
              "effort": [
                "none"
              ]
            }
          },
          "compatibility": {
            "reasoningField": "reasoning_content",
            "maxTokensField": "max_tokens",
            "streamUsage": true
          }
        }
      }
    }
  },
  "model": "vora/qwen3.8-27b"
}

The configuration includes context and output limits for each model. To start with Qwen3.8 27B directly:

opencode --model vora/qwen3.8-27b

If a desktop launcher cannot see VORA_API_KEY, run /connect and save your Vora credential for provider ID vora. Keep the endpoint and model definitions in the configuration. For long-context models, the generated provider settings extend both the response-header and total timeout; keep those settings when merging configurations.

OpenCode provider documentation and timeout settings.

Cline

Open Cline's provider settings and enter:

  1. API Provider: OpenAI Compatible.
  2. Base URL: https://api.voracompute.com/v1.
  3. API Key: your Vora key.
  4. Model ID: a model ID from the table above, such as qwen3.8-27b.
  5. Context Window and Max Output Tokens: the values for that same model in the table.

Save the settings and start a new task.

Cline provider documentation.

Continue

  1. Open your local configuration from Continue's settings (~/.continue/config.yaml).
  2. Add VORA_API_KEY=vora_sk_..., with your actual key, to ~/.continue/.env in your home directory.
  3. Add these entries to your configuration. If you already have a models list, append the model entries to it.
  4. Restart your IDE after saving the key, then select a Vora model in Continue. Use Agent mode when you want it to call tools.

On Windows, the same files live in %USERPROFILE%\.continue\. See Continue's secrets guide if the key is not picked up.

name: My Config
version: 0.0.1
schema: v1
models:
  - name: Qwen3.8 27B (Vora)
    provider: openai
    model: qwen3.8-27b
    apiBase: https://api.voracompute.com/v1
    apiKey: ${{ secrets.VORA_API_KEY }}
    useResponsesApi: false
    roles: [chat, edit, apply]
    capabilities: [tool_use]
    defaultCompletionOptions:
      contextLength: 1000000
      maxTokens: 8192
  - name: Nemotron 3.5 Super VL (Vora)
    provider: openai
    model: nemotron-3.5-super-vl-120b
    apiBase: https://api.voracompute.com/v1
    apiKey: ${{ secrets.VORA_API_KEY }}
    useResponsesApi: false
    roles: [chat, edit, apply]
    capabilities: [tool_use]
    defaultCompletionOptions:
      contextLength: 1000000
      maxTokens: 8192
  - name: EuroLLM 22B (Vora)
    provider: openai
    model: eurollm-22b
    apiBase: https://api.voracompute.com/v1
    apiKey: ${{ secrets.VORA_API_KEY }}
    useResponsesApi: false
    roles: [chat, edit, apply]
    capabilities: [tool_use]
    defaultCompletionOptions:
      contextLength: 32768
      maxTokens: 4096

The configuration uses Chat Completions, enables tool calling and sets each model's limits. The examples configure text tasks; they do not enable image attachments.

Continue configuration reference and OpenAI-compatible provider settings.

Aider

Set VORA_API_KEY first. Save the following as .aider.model.settings.yml in your project, merging with existing entries if needed. It sets the output limit for each Vora model.

- name: openai/qwen3.8-27b
  extra_params:
    max_tokens: 8192
- name: openai/nemotron-3.5-super-vl-120b
  extra_params:
    max_tokens: 8192
- name: openai/eurollm-22b
  extra_params:
    max_tokens: 4096

Then run Aider from that project directory:

export OPENAI_API_BASE="https://api.voracompute.com/v1"
export OPENAI_API_KEY="$VORA_API_KEY"
aider --model openai/qwen3.8-27b

To choose another model, replace the ID after openai/ with one from the table. Aider may warn that a custom model is unknown. Keep the openai/ prefix and add only the files needed for your task. These environment variables apply to this terminal session.

Aider OpenAI-compatible setup and custom model settings.

Cherry Studio

Add a generic OpenAI-compatible provider using Chat Completions, the base URL above and your Vora API key. Add a model ID from the table, then select it in a chat. If the app asks for context or output limits, copy the matching values.

For Qwen3.8 27B, use Off, Low, Medium or High. Default leaves the choice to Vora, which uses that model's default level. The request must send reasoning_effort; an accompanying top-level enable_thinking boolean does not override it. Other models have the effort levels listed in the table.

Cherry Studio 2.1.7 was checked for these five request settings with the default model configuration. Its slider supports arrow keys, Home for Off and End for High. Versions and model catalogs can expose different controls. If levels are missing, set reasoning_effort as a custom request parameter, or use Chat. Remove fixed custom effort and legacy thinking overrides before using the slider so Default can follow Vora's default.

Codex

Requires an adapter. Vora integration not yet validated in the Codex app or CLI.

Codex providers use the Responses API (/v1/responses). Vora currently serves Chat Completions (/v1/chat/completions), so entering the Vora base URL and key alone will not connect it.

An adapter would need to translate requests, streaming events, tool calls and conversation history between those APIs. Vora does not yet provide a tested adapter or a copy-and-run Codex configuration. A model appearing in the picker does not establish that agent tasks work.

To use Vora for coding now, follow the OpenCode or Cline setup above. Check the official Codex provider configuration for its protocol requirements.

Codex can also delegate tasks to a Vora model with the Vora CLI.

Claude Desktop and Claude Code

No direct connection to the current Vora API.

Claude Desktop's third-party inference gateway requires the Anthropic Messages API (/v1/messages), including streaming and tool use. Claude Code also expects an Anthropic-compatible gateway. Vora currently exposes Chat Completions, so its base URL cannot be used directly in those gateway settings. Routing Vora models through an adapter would need separate compatibility testing; Vora has not validated it.

Claude Desktop can also call external tools through MCP connectors. A Vora MCP connector could let Claude request an answer from a Vora model as a tool, while Claude remains the main assistant. Vora does not currently provide that connector, and the API base URL is not an MCP server URL.

Claude Code can delegate tasks to a Vora model with the Vora CLI, while Claude remains the main assistant.

For a desktop chat app that connects to Vora directly, use Cherry Studio. See Claude Desktop gateway requirements, Claude Code gateways and Claude's MCP connectors.

3. Check your connection

Start a new conversation with your selected Vora model and ask: “Say hello in one sentence.” Then, in a coding agent, ask it to list the files in your current project without changing them. Review the tool request and confirm that the result returns to the conversation.

If the tool cannot connect, try this independent API check from a macOS or Linux terminal with VORA_API_KEY set:

curl --fail-with-body https://api.voracompute.com/v1/chat/completions \
  -H "Authorization: Bearer $VORA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "qwen3.8-27b",
  "messages": [
    {
      "role": "user",
      "content": "Say hello in one sentence."
    }
  ],
  "reasoning_effort": "none",
  "max_tokens": 128
}'

A successful response contains choices[0].message.content. This checks the key and API; the small task in your agent checks its tool flow. You can also try the selected model in Chat.

If something goes wrong

Optional reasoning

Every chat model defaults to Medium. If your client exposes request parameters, set reasoning_effort to one of the values in that model's row. minimal is accepted as Low, and xhigh or max as High. Use none for direct answers. Higher effort uses output tokens and can take longer; it does not guarantee a better answer.

A client can display reasoning_content separately from the answer. See Thinking modes for legacy settings, token accounting and cancellation.

LangChain

Install langchain-openai and set VORA_API_KEY in your terminal. To use another model, replace the model ID with one from the table above.

import os
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    base_url="https://api.voracompute.com/v1",
    api_key=os.environ["VORA_API_KEY"],
    model="qwen3.8-27b",
)

LlamaIndex

Install llama-index-llms-openai-like and set VORA_API_KEY. If you switch models, update both the model ID and context window from the table.

import os
from llama_index.llms.openai_like import OpenAILike

llm = OpenAILike(
    model="qwen3.8-27b",
    api_base="https://api.voracompute.com/v1",
    api_key=os.environ["VORA_API_KEY"],
    context_window=1000000,
    is_chat_model=True,
    is_function_calling_model=True,
)

Vercel AI SDK

Install ai and @ai-sdk/openai-compatible, and set VORA_API_KEY in your server environment. Keep the key on the server. To use another model, replace the ID with one from the table.

import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
import { generateText } from "ai";

const vora = createOpenAICompatible({
  name: "vora",
  baseURL: "https://api.voracompute.com/v1",
  apiKey: process.env.VORA_API_KEY,
});

const { text } = await generateText({ model: vora("qwen3.8-27b"), prompt: "Hello" });