Providers

Provider Setup Guides

Model naming, detection, and configuration quirks for each platform Hyperflow supports. See model-routing for how Hyperflow selects between thinking and worker tiers at runtime, and installation for the initial setup flow.

Claude Code

Detection: CLAUDE_CODE_* environment variables, set automatically by the CLI.

Default models:

Claude Code

Thinking Opus 4.8

Worker Sonnet 4.6

Available models:

Model ID	Name	Tier	Notes
`opus-4-8`	Opus 4.8	Thinking	Latest — Hyperflow default
`opus-4-7`	Opus 4.7	Thinking	Previous Opus
`opus-4-6`	Opus 4.6	Thinking	Legacy
`opus-4-5`	Opus 4.5	Thinking	Legacy
`sonnet-4-6`	Sonnet 4.6	Worker	Hyperflow default
`sonnet-4-5`	Sonnet 4.5	Worker	Legacy
`haiku-4-5`	Haiku 4.5	Worker	Fast and cheap

Version pinning: Claude Code's Agent tool accepts "opus", "sonnet", "haiku" aliases that resolve to the latest version. To pin a specific version:

Version pinning

export ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-4-8
export ANTHROPIC_DEFAULT_SONNET_MODEL=claude-sonnet-4-6

Dynamic detection: Hyperflow reads ~/.claude/settings.json to detect the currently configured model.

OpenCode

Detection: OPENCODE_* environment variables or opencode binary in PATH.

Default models:

OpenCode

Thinking Claude Opus 4.8

Worker Claude Sonnet 4.6

Available models:

Model ID	Name	Provider	Tier	Notes
`anthropic/claude-opus-4-8`	Claude Opus 4.8	Anthropic	Thinking	Hyperflow default
`anthropic/claude-opus-4-7`	Claude Opus 4.7	Anthropic	Thinking	Previous Opus
`anthropic/claude-opus-4-6`	Claude Opus 4.6	Anthropic	Thinking	Legacy
`openai/gpt-5.5`	GPT-5.5	OpenAI	Thinking	Latest GPT
`openai/gpt-5.4`	GPT-5.4	OpenAI	Thinking	Cached input discount
`deepseek/deepseek-v4-pro`	DeepSeek V4 Pro	DeepSeek	Thinking	Open-weight
`anthropic/claude-sonnet-4-6`	Claude Sonnet 4.6	Anthropic	Worker	Hyperflow default
`anthropic/claude-haiku-4-5`	Claude Haiku 4.5	Anthropic	Worker	Fast and cheap
`openai/gpt-5.4-mini`	GPT-5.4 Mini	OpenAI	Worker	Cost-efficient

Model format: OpenCode uses provider/model format — e.g. anthropic/claude-opus-4-8.

75+ providers: OpenCode supports Anthropic, OpenAI, Amazon Bedrock, Azure, DeepSeek, xAI, Ollama, LM Studio, OpenRouter, Together AI, Groq, and many more via the Vercel AI SDK.

Dynamic detection: Hyperflow runs opencode models list --json (2s timeout) to get the full list of available models. This captures BYOK and custom provider models.

OpenCode config: Models can also be set in opencode.json via "model" and "small_model" fields.

Codex

Detection: CODEX_* environment variables or the codex binary in PATH.

Default models:

Codex

Thinking GPT-5.5 · adaptive

Worker GPT-5.4 · fast

Available models:

Model ID	Name	Provider	Tier	Notes
`gpt-5.5`	GPT-5.5	OpenAI	Thinking	Codex default for orchestrator, reviewer, debugger, decision-maker, brainstormer
`gpt-5.4`	GPT-5.4	OpenAI	Worker	Codex fast worker default; can be used as thinking fallback
`gpt-5.4-mini`	GPT-5.4 Mini	OpenAI	Worker	Lower-cost worker fallback

Reasoning policy: Thinking roles use task-adaptive reasoning: low for trivial docs/config checks, medium for normal planning/review, and high for debugging, architecture, security, and final integration. Worker roles stay on low reasoning for fast mode unless explicitly overridden. Codex defaults never use xhigh.

Installing into Codex:

Codex

codex plugin marketplace add Mohammed-Abdelhady/hyperflow
codex plugin add hyperflow@hyperflow-marketplace

Codex reads project instructions from AGENTS.md; run scripts/setup-detection.sh --tools codex <project> to generate the project shim.

Antigravity

Antigravity is Google's agent-first IDE — a VS Code fork announced November 2025 alongside Gemini 3. It runs agents on Gemini natively and also supports Claude and the open-weights GPT-OSS model.

Detection: ANTIGRAVITY_* environment variables or antigravity binary in PATH.

Default models:

Antigravity

Thinking Gemini 3 Pro

Worker Gemini 3.5 Flash

Available models:

Model ID	Name	Provider	Tier	Notes
`gemini-3-pro`	Gemini 3 Pro	Google	Thinking	Antigravity default
`gemini-3.1-pro`	Gemini 3.1 Pro	Google	Thinking	2M context window
`claude-opus-4-8`	Claude Opus 4.8	Anthropic	Thinking	Antigravity also supports Claude
`gpt-oss-120b`	GPT-OSS-120B	OpenAI	Thinking	Open-weights variant
`gemini-3.5-flash`	Gemini 3.5 Flash	Google	Worker	Antigravity default
`gemini-3-flash`	Gemini 3 Flash	Google	Worker	Stable Flash
`claude-sonnet-4-6`	Claude Sonnet 4.6	Anthropic	Worker	Antigravity also supports Claude

Model format: Model IDs are bare slugs — e.g. gemini-3-pro. The per-agent model picker in Antigravity's Manager surface shows display names.

Dynamic detection: Hyperflow reads ~/.gemini/antigravity/settings.json to detect the configured model.

Installing into Antigravity: Antigravity loads agent skills from ~/.gemini/antigravity/skills/ as directory packages with a SKILL.md — the same format Hyperflow already uses. Symlink or copy Hyperflow's skills/* into that directory and they're available across every Antigravity project.

Antigravity also merges a global ~/.gemini/AGENTS.md and a per-workspace AGENTS.md at session start — the Hyperflow bridge doctrine subset can live there (the AGENTS.md analogue of CLAUDE.md).