Providers and Models

Perstack supports multiple LLM providers. Configure via CLI options, environment variables, or perstack.toml.

Default Model

Perstack uses claude-sonnet-4-5 as the default model, selected based on:

Standard tier pricing — not a flagship model, economically sustainable for extended runs
High agentic performance — demonstrated tool use capability in benchmarks like 𝜏²-Bench

The default balances cost efficiency with reliable agent behavior. As new models are released, the default may change based on these criteria.

To override the default, specify in perstack.toml:

model = "gemini-2.5-pro"

[provider]
providerName = "google"

Or via CLI:

npx perstack run my-expert "query" --provider google --model gemini-2.5-pro

Supported Providers

Provider	Key	Description
Anthropic	`anthropic`	Claude models (default)
Google	`google`	Gemini models
OpenAI	`openai`	GPT and reasoning models
DeepSeek	`deepseek`	DeepSeek models
Ollama	`ollama`	Local model hosting
Azure OpenAI	`azure-openai`	Azure-hosted OpenAI models
Amazon Bedrock	`amazon-bedrock`	AWS Bedrock-hosted models
Google Vertex AI	`google-vertex`	Google Cloud Vertex AI models

Anthropic

Environment variables:

Variable	Required	Description
`ANTHROPIC_API_KEY`	Yes	API key
`ANTHROPIC_BASE_URL`	No	Custom endpoint

perstack.toml settings:

[provider]
providerName = "anthropic"
[provider.setting]
baseUrl = "https://custom-endpoint.example.com"  # Optional
headers = { "X-Custom-Header" = "value" }        # Optional

Setting	Type	Description
`baseUrl`	string	Custom API endpoint
`headers`	object	Custom HTTP headers

Native reasoning: Supported via extended thinking.

Models:

Model	Context	Max Output
`claude-opus-4-5`	200K	32K
`claude-opus-4-1`	200K	32K
`claude-opus-4-20250514`	200K	32K
`claude-sonnet-4-5`	200K	64K
`claude-sonnet-4-20250514`	200K	64K
`claude-3-7-sonnet-20250219`	200K	64K
`claude-haiku-4-5`	200K	8K
`claude-3-5-haiku-latest`	200K	8K

export ANTHROPIC_API_KEY=sk-ant-...
npx perstack run my-expert "query" --provider anthropic --model claude-sonnet-4-5

Google

Environment variables:

Variable	Required	Description
`GOOGLE_GENERATIVE_AI_API_KEY`	Yes	API key
`GOOGLE_GENERATIVE_AI_BASE_URL`	No	Custom endpoint

perstack.toml settings:

[provider]
providerName = "google"
[provider.setting]
baseUrl = "https://custom-endpoint.example.com"  # Optional
headers = { "X-Custom-Header" = "value" }        # Optional

Setting	Type	Description
`baseUrl`	string	Custom API endpoint
`headers`	object	Custom HTTP headers

Models:

Model	Context	Max Output
`gemini-3-pro-preview`	1M	64K
`gemini-2.5-pro`	1M	64K
`gemini-2.5-flash`	1M	64K
`gemini-2.5-flash-lite`	1M	64K

export GOOGLE_GENERATIVE_AI_API_KEY=AIza...
npx perstack run my-expert "query" --provider google --model gemini-2.5-pro

OpenAI

Environment variables:

Variable	Required	Description
`OPENAI_API_KEY`	Yes	API key
`OPENAI_BASE_URL`	No	Custom endpoint (OpenAI-compatible)
`OPENAI_ORGANIZATION`	No	Organization ID
`OPENAI_PROJECT`	No	Project ID

perstack.toml settings:

[provider]
providerName = "openai"
[provider.setting]
baseUrl = "https://custom-endpoint.example.com"  # Optional
organization = "org-xxx"                          # Optional
project = "proj-xxx"                              # Optional
name = "custom-openai"                            # Optional: custom provider name
headers = { "X-Custom-Header" = "value" }         # Optional

Setting	Type	Description
`baseUrl`	string	Custom API endpoint
`organization`	string	OpenAI organization ID
`project`	string	OpenAI project ID
`name`	string	Custom provider name
`headers`	object	Custom HTTP headers

Native reasoning: Supported via reasoningEffort. Works with o-series models (o1, o3, o4-mini).

Models:

Model	Context	Max Output
`gpt-5`	400K	128K
`gpt-5-mini`	400K	128K
`gpt-5-nano`	400K	128K
`gpt-5-chat-latest`	128K	16K
`o4-mini`	200K	100K
`o3`	200K	10K
`o3-mini`	200K	10K
`gpt-4.1`	1M	32K

export OPENAI_API_KEY=sk-proj-...
npx perstack run my-expert "query" --provider openai --model gpt-5

DeepSeek

Environment variables:

Variable	Required	Description
`DEEPSEEK_API_KEY`	Yes	API key
`DEEPSEEK_BASE_URL`	No	Custom endpoint

perstack.toml settings:

[provider]
providerName = "deepseek"
[provider.setting]
baseUrl = "https://custom-endpoint.example.com"  # Optional
headers = { "X-Custom-Header" = "value" }        # Optional

Setting	Type	Description
`baseUrl`	string	Custom API endpoint
`headers`	object	Custom HTTP headers

Models:

Model	Context	Max Output
`deepseek-chat`	128K	8K
`deepseek-reasoner`	128K	64K

export DEEPSEEK_API_KEY=sk-...
npx perstack run my-expert "query" --provider deepseek --model deepseek-chat

Ollama

Environment variables:

Variable	Required	Description
`OLLAMA_BASE_URL`	No	Server URL (default: `http://localhost:11434`)

perstack.toml settings:

[provider]
providerName = "ollama"
[provider.setting]
baseUrl = "http://localhost:11434"        # Optional
headers = { "X-Custom-Header" = "value" } # Optional

Setting	Type	Description
`baseUrl`	string	Ollama server URL
`headers`	object	Custom HTTP headers

Models:

Model	Context	Max Output
`gpt-oss:20b`	128K	128K
`gpt-oss:120b`	128K	128K
`gemma3:1b`	32K	32K
`gemma3:4b`	128K	128K
`gemma3:12b`	128K	128K
`gemma3:27b`	128K	128K

export OLLAMA_BASE_URL=http://localhost:11434
npx perstack run my-expert "query" --provider ollama --model gpt-oss:20b

Azure OpenAI

Environment variables:

Variable	Required	Description
`AZURE_API_KEY`	Yes	API key
`AZURE_RESOURCE_NAME`	Yes	Resource name
`AZURE_API_VERSION`	No	API version
`AZURE_BASE_URL`	No	Custom endpoint

perstack.toml settings:

[provider]
providerName = "azure-openai"
[provider.setting]
resourceName = "your-resource-name"              # Optional (env fallback)
apiVersion = "2024-02-15-preview"                # Optional
baseUrl = "https://custom-endpoint.example.com"  # Optional
headers = { "X-Custom-Header" = "value" }        # Optional
useDeploymentBasedUrls = true                    # Optional

Setting	Type	Description
`resourceName`	string	Azure resource name
`apiVersion`	string	Azure API version
`baseUrl`	string	Custom API endpoint
`headers`	object	Custom HTTP headers
`useDeploymentBasedUrls`	boolean	Use deployment-based URLs

export AZURE_API_KEY=your_azure_key
export AZURE_RESOURCE_NAME=your_resource_name
npx perstack run my-expert "query" --provider azure-openai --model your-deployment-name

Amazon Bedrock

Environment variables:

Variable	Required	Description
`AWS_ACCESS_KEY_ID`	Yes	Access key ID
`AWS_SECRET_ACCESS_KEY`	Yes	Secret access key
`AWS_REGION`	Yes	Region (e.g., `us-east-1`)
`AWS_SESSION_TOKEN`	No	Session token (temporary credentials)

perstack.toml settings:

[provider]
providerName = "amazon-bedrock"
[provider.setting]
region = "us-east-1"  # Optional (env fallback)

Setting	Type	Description
`region`	string	AWS region

export AWS_ACCESS_KEY_ID=AKIA...
export AWS_SECRET_ACCESS_KEY=...
export AWS_REGION=us-east-1
npx perstack run my-expert "query" --provider amazon-bedrock --model anthropic.claude-v2

Google Vertex AI

Environment variables:

Variable	Required	Description
`GOOGLE_VERTEX_PROJECT`	No	GCP project ID
`GOOGLE_VERTEX_LOCATION`	No	GCP location (e.g., `us-central1`)
`GOOGLE_VERTEX_BASE_URL`	No	Custom endpoint

perstack.toml settings:

[provider]
providerName = "google-vertex"
[provider.setting]
project = "my-gcp-project"                       # Optional
location = "us-central1"                         # Optional
baseUrl = "https://custom-endpoint.example.com"  # Optional
headers = { "X-Custom-Header" = "value" }        # Optional

Setting	Type	Description
`project`	string	GCP project ID
`location`	string	GCP location
`baseUrl`	string	Custom API endpoint
`headers`	object	Custom HTTP headers

export GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
npx perstack run my-expert "query" --provider google-vertex --model gemini-1.5-pro