Provider Guides

June 5, 2026 · View on GitHub

This index collects provider-specific guides for configuring VT Code with different LLM backends.

Google Gemini

OpenAI GPT

  • Official docs:
  • Follow the Getting Started guide for API key setup.
  • See vtcode-core/src/config/constants.rs for the latest supported models.
  • GPT-5.2 reference: Using GPT-5.2
  • VT Code's default OpenAI profile is gpt-5.4 with reasoning_effort = "none" and verbosity = "medium"; raise reasoning only when the task shape justifies the extra latency.
  • VT Code applies a compact GPT-5.4 prompt contract rather than a verbatim cookbook prompt: compact outputs, low-risk follow-through, dependency-aware tool use, completeness checks, verification, and conditional grounding/citation rules.
  • File inputs are supported for native OpenAI Responses API requests through input_file parts.
  • Supported file input fields in VT Code message parts: file_id, file_data, file_url, filename.
  • file_url is Responses API only; VT Code rejects file_url when a request uses Chat Completions.
  • VT Code only upgrades local non-image file refs such as @report.pdf and @"Quarterly Deck.pptx" into structured file attachments for native OpenAI Responses sessions on api.openai.com.
  • Remote external document URLs such as @https://example.com/letter.pdf are also only elevated to structured file_url inputs for native OpenAI Responses sessions.
  • ChatGPT subscription sessions, OpenAI-compatible endpoints, and other providers keep non-image @file refs as plain text plus file-reference metadata so the agent can resolve the path and read the file with tools.
  • Raw image paths still use the existing multimodal image path flow. Non-image files require explicit @... references.
  • Official OpenAI Responses replays now preserve assistant phase metadata for replayed assistant history (commentary for preambles/progress updates, final_answer for completed answers) when the target GPT model supports it. VT Code does not send this field to Chat Completions, tool/user items, or non-native OpenAI-compatible endpoints.
  • OpenAI Responses hosted tools currently map through ToolDefinition for web_search, file_search, hosted tool_search, and remote mcp, with hosted config passed through directly on each tool entry.
  • OpenAI hosted shell mounts are configured through provider.openai.hosted_shell in vtcode.toml.
  • Hosted shell skill mounts support both skill_reference and inline bundle entries; VT Code forwards them to OpenAI but does not upload/create hosted skills in this path.
  • This hosted-shell workflow is separate from VT Code's local SKILL.md filesystem skills.
  • For large corpora, prefer File Search/Retrieval instead of sending full files inline.
  • For spreadsheet-heavy analysis, use Hosted Shell workflows instead of large inline sheet prompts.

Anthropic Claude

GitHub Copilot

  • Guide: GitHub Copilot Managed Auth
  • Runtime dependency: copilot must be installed and runnable for login/logout
  • Optional fallback: gh is only used when VT Code probes an existing GitHub CLI auth session
  • Commands: vtcode login copilot, vtcode logout copilot, /login copilot, /logout copilot

OpenRouter Marketplace

  • Guide: OpenRouter Integration
  • Official docs:
  • Default model: xiaomi/mimo-v2.5-pro (VT Code's default). Xiaomi MiMo V2.5 and V2.5 Pro are also available.
  • Xiaomi MiMo models:
    • xiaomi/mimo-v2.5-pro — flagship agentic model, 1M context, reasoning + tool calls
    • xiaomi/mimo-v2.5 — omnimodal model, 1M context, reasoning + tool calls

Atlas Cloud

  • Guide: Atlas Cloud Integration
  • Official docs:
  • Integration mode: configure Atlas Cloud through VT Code's [[custom_providers]] support because the LLM endpoint is OpenAI-compatible.
  • Base URL: https://api.atlascloud.ai/v1
  • Recommended model: start with deepseek-ai/deepseek-v4-flash (DeepSeek's latest flash model, 1M context, $0.14/M input tokens).

Xiaomi MiMo

  • Provider key: mimo
  • Docs: Xiaomi MiMo Platform
  • Pricing: Pay-as-you-go · Subscription · Quick Access
  • Setup: Set MIMO_API_KEY or use the MiMo provider in VT Code's configuration
  • Models:
    • mimo-v2.5-pro — flagship agentic model, 1M context, deep thinking
    • mimo-v2.5 — omnimodal model (text, image, audio, video), 1M context

Ollama Local & Cloud Models

  • Guide: Local Inference Servers (unified /local command)
  • Setup: Install and run Ollama locally (official install)
  • Configuration: Local usage needs no key; set OLLAMA_API_KEY to access Ollama Cloud
  • Default model: Any locally available model (e.g., llama3:8b, mistral:7b, qwen3:1.7b)
  • Cloud models: Use IDs like gpt-oss:120b-cloud with OLLAMA_BASE_URL=https://ollama.com
  • Custom Models: Use the custom-ollama option in the model picker to enter any locally or cloud-available model ID
  • Base URL: Configurable via OLLAMA_BASE_URL environment variable (defaults to http://localhost:11434)
  • Features: Streaming, structured tool calling (including Ollama's web search tools), and thinking traces when reasoning_effort is enabled

LM Studio Local Server

  • Guide: LM Studio Provider Guide
  • Server: Enable the OpenAI-compatible Developer server in LM Studio (defaults to http://localhost:1234/v1)
  • Environment: Optional LMSTUDIO_API_KEY when auth is enabled; override host/port via LMSTUDIO_BASE_URL
  • Default model: lmstudio-community/meta-llama-3.1-8b-instruct (local inference)
  • Catalog: Also ships with lmstudio-community/meta-llama-3-8b-instruct, lmstudio-community/qwen2.5-7b-instruct, lmstudio-community/gemma-2-2b-it, lmstudio-community/gemma-2-9b-it, and lmstudio-community/phi-3.1-mini-4k-instruct, plus any custom GGUF models you expose
  • Features: Streaming, tool calling, structured output, and reasoning effort passthrough via the shared OpenAI surface

llama.cpp Local Server

  • Guide: llama.cpp Provider Guide
  • Server: VT Code targets llama-server and defaults to http://localhost:8080/v1
  • Environment: LLAMACPP_BASE_URL overrides the endpoint; LLAMACPP_MODEL_PATH enables VT Code-managed startup
  • Managed startup: VT Code can launch llama-server -m /path/to/model.gguf --port ... when the endpoint is localhost and a GGUF path is configured
  • Starter catalog: gpt-oss-20b, qwen3.6-27b, qwen3.6-35b-a3b, gemma-4-26b-a4b, gemma-4-e4b, and step-3.5-flash
  • Features: Streaming, dynamic /v1/models discovery, local no-auth defaults, and OpenAI-compatible request handling
  • Provider key: evolink
  • Official docs: Evolink Docs
  • Base URL: https://direct.evolink.ai/v1
  • Auth: EVOLINK_API_KEY environment variable
  • Setup: Set EVOLINK_API_KEY from Evolink dashboard, then configure provider = "evolink" in vtcode.toml
  • Models:
    • evolink/gpt-5.2 (default)
    • evolink/gpt-5.5
    • evolink/deepseek-v4-pro
    • evolink/deepseek-v4-flash
    • evolink/doubao-seed-2.0-pro
    • evolink/gemini-3.1-pro-preview
    • evolink/gemini-3.5-flash
    • evolink/MiniMax-M3
    • evolink/claude-sonnet-4-6
    • evolink/claude-opus-4-8
    • evolink/claude-haiku-4-5-20251001
  • Features: OpenAI-compatible gateway exposing many upstream models behind one endpoint. Evolink serves models under bare upstream names (e.g. gpt-5.2) that collide with VT Code's first-class providers, so curated model IDs are namespaced as evolink/<model>. The provider strips the prefix before sending requests upstream.

Anthropic API Compatibility Server

VT Code provides compatibility with the Anthropic Messages API to help connect existing applications to VT Code, including tools like Claude Code.

  • Feature: Anthropic API compatibility server
  • Command: vtcode anthropic-api --port 11434
  • Endpoint: /v1/messages (mirrors Anthropic Messages API)
  • Environment variables:
    • ANTHROPIC_AUTH_TOKEN=ollama (required but ignored)
    • ANTHROPIC_BASE_URL=http://localhost:11434
    • ANTHROPIC_API_KEY=ollama (required but ignored)
  • Features: Streaming, tool calling, vision support, multi-turn conversations

ℹ Additional provider-specific guides will be added as new integrations land in VT Code.