SDK Support

Netclode supports multiple coding agent SDKs. Pick the one that fits your workflow.

Overview

SDK	Provider	Authentication	Key Features
Claude Agent	Anthropic	API key	Extended thinking, native tools, session persistence
OpenCode	Multi-provider	API keys	Anthropic, OpenAI, Mistral support
Copilot	GitHub / Anthropic	GitHub token or API key	Premium quota tracking, billing multipliers
Codex	OpenAI	API key or ChatGPT OAuth	Reasoning effort levels, thread persistence

Claude Agent SDK (Default)

Direct integration with Anthropic's Claude via @anthropic-ai/claude-agent-sdk.

Auth: ANTHROPIC_API_KEY

Models:

• claude-opus-4-5-20251101 (default) - most capable, extended thinking
• claude-sonnet-4-0 - balanced

Features: Extended thinking, native tools, session persistence, interrupt support.

Configuration

# .env
ANTHROPIC_API_KEY=sk-ant-api03-xxx

OpenCode SDK

Multi-provider support through the OpenCode CLI in server mode.

Auth: Set whichever API keys you need:

ANTHROPIC_API_KEY=sk-ant-xxx
OPENAI_API_KEY=sk-xxx
MISTRAL_API_KEY=xxx
ZAI_API_KEY=xxx  # For Z.AI GLM-4.7 models

Models: Format is provider/model-name:

• anthropic/claude-sonnet-4-0 (default)
• anthropic/claude-sonnet-4-5-20250514 (with thinking)
• openai/gpt-4o
• mistral/mistral-large-latest
• zai/GLM-4.7 (via Z.AI - cheaper alternative)

Thinking budgets: high (16k tokens), max (32k tokens)

GitHub Copilot SDK

Uses @github/copilot-sdk with two backend options.

GitHub Backend

Uses your Copilot subscription. Create a fine-grained PAT at https://github.com/settings/tokens?type=beta with Copilot read-only permission.

GITHUB_COPILOT_TOKEN=github_pat_xxx

Includes quota tracking and billing multipliers (0.33x to 50x depending on model).

Anthropic Backend (BYOK)

Use Anthropic API directly without GitHub subscription:

ANTHROPIC_API_KEY=sk-ant-xxx

OpenAI Codex SDK

Uses @openai/codex-sdk.

API Key Mode

OPENAI_API_KEY=sk-xxx

ChatGPT OAuth Mode

Use your ChatGPT subscription instead of API credits:

netclode auth codex

Opens browser flow and outputs tokens for your .env.

Models

• codex-mini-latest - fast
• gpt-5-codex - most capable

Reasoning Effort

Controls how much "thinking" the model does: minimal, low, medium (default), high, xhigh. Higher = better quality but more latency.

Session Configuration

Specify SDK and model when creating a session:

netclode sessions create --repo owner/repo --sdk claude
netclode sessions create --repo owner/repo --sdk opencode --model anthropic/claude-sonnet-4-5-20250514
netclode sessions create --repo owner/repo --sdk copilot
netclode sessions create --repo owner/repo --sdk codex --model codex-mini-latest

Or use the iOS app model picker.

Environment Variables Reference

Variable	SDK	Description
ANTHROPIC_API_KEY	Claude, OpenCode, Copilot (BYOK)	Anthropic API key
OPENAI_API_KEY	OpenCode, Codex	OpenAI API key
MISTRAL_API_KEY	OpenCode	Mistral API key
ZAI_API_KEY	OpenCode	Z.AI API key (for GLM-4.7 models)
GITHUB_COPILOT_TOKEN	Copilot	GitHub PAT with copilot scope
CODEX_ACCESS_TOKEN	Codex	ChatGPT OAuth access token
CODEX_ID_TOKEN	Codex	ChatGPT OAuth ID token
CODEX_REFRESH_TOKEN	Codex	ChatGPT OAuth refresh token

Local Models with Ollama

Run local LLMs with GPU acceleration. Requires NVIDIA GPU and Ollama deployment.

Setup

cd infra/ansible
DEPLOY_HOST=your-server NVIDIA_ENABLED=true OLLAMA_ENABLED=true \
  ansible-playbook playbooks/site.yaml

See infra/ansible/README.md for full setup.

Usage

# Pull a model
kubectl --context netclode -n netclode exec -it deploy/ollama -- ollama pull qwen2.5-coder:32b

# Use in session
netclode sessions create --repo owner/repo --sdk opencode --model ollama/qwen2.5:7b-instruct

Models show up in the iOS app picker once Ollama is running.

Recommended models (16GB VRAM)

• qwen2.5-coder:32b-instruct-q4_K_M - best coding
• deepseek-coder-v2:16b - fast coding
• mistral:7b-instruct - fast general