Providers¶

Perspt supports multiple LLM providers through the genai Rust client crate, which provides unified access to all major commercial, open-source, and cloud-provider model APIs.

Supported Providers and Adapters¶

Perspt accepts nine provider identifiers, each routed to the matching genai adapter (any other value is rejected as unsupported):

Provider / Adapter	Adapter Kind ID	Reference Model	Notes
OpenAI	`openai`	`gpt-5.5`	SOTA GPT-5.5, GPT-5-mini
Anthropic	`anthropic`	`claude-fable`	Claude Fable, Opus 4.8, Sonnet 4.6, Haiku 4.6
Google Gemini	`gemini`, `google`	`gemini-3.5-flash`	Gemini 3.5 Flash, 3.1 Pro, 3.1 Flash-Lite
Google Vertex AI	`vertex`	`vertex::gemini-3.5-flash`	Google Cloud Vertex platform
Groq	`groq`	`llama-3.3-70b-specdec`	Ultra-low latency Llama/Gemma on LPU
Cohere	`cohere`	`command-a-plus`	Command A+, North Mini Code
xAI	`xai`	`grok-4`	Grok 4 family
DeepSeek	`deepseek`	`deepseek-v4`	DeepSeek v4 models (Chat, Coder)
Ollama	`ollama`	`llama3.3`	Local offline models

Configuration Methods¶

1. Environment Variables (recommended):

export GEMINI_API_KEY="your-key"
perspt

2. CLI Flags:

perspt chat --model gpt-5.5

3. Config File (config.toml):

provider = "anthropic"
model = "claude-fable"

Agent mode can additionally bind several credentials and routes at once through the [providers.<id>] and [models] tables, which hold fully qualified provider::model routes for multi-route portfolios. See Agent Mode for details.

Provider-Specific Notes¶

OpenAI

export OPENAI_API_KEY="sk-xxx"
perspt chat --model gpt-5.5

Azure OpenAI (via OpenAI Compatible)

Azure OpenAI requires configuring the base URL override and the API key:

export OPENAI_API_KEY="your-azure-key"
export OPENAI_BASE_URL="https://your-resource.openai.azure.com/openai/deployments/your-deployment"
perspt chat --model gpt-5.5

Anthropic

export ANTHROPIC_API_KEY="sk-ant-xxx"
perspt chat --model claude-fable

Google Gemini

export GEMINI_API_KEY="AIza..."
perspt chat --model gemini-3.5-flash

Google Vertex AI

Vertex AI requires your Google Cloud project ID and a location (optional, defaults to global). Authentication is typically handled via Google Application Default Credentials (ADC); setting VERTEX_API_KEY to a bearer token overrides the ADC token.

export VERTEX_PROJECT_ID="my-gcp-project-123"
export VERTEX_LOCATION="us-central1"
# Run using Vertex model prefix
perspt chat --model vertex::gemini-3.5-flash

Ollama (Local)

ollama serve
ollama pull llama3.3
perspt chat --model llama3.3

No API key required. Perspt auto-detects Ollama as the fallback provider.