Providers

Hamr supports OpenAI-compatible and Anthropic Messages providers. You can configure multiple providers and switch between them at runtime.

Provider Protocols

Protocol	Description
openai-compatible	Any service with an OpenAI-compatible /chat/completions endpoint
anthropic-messages	Real Anthropic Messages API adapter via /v1/messages

Built-in Provider Presets

Hamr ships with presets for these providers:

Provider ID	Protocol	Base URL	Auth
relay	openai-compatible	http://127.0.0.1:1234/v1	none
custom	openai-compatible	user-configured	optional
deepseek	openai-compatible	https://api.deepseek.com/v1	DEEPSEEK_API_KEY
openrouter	openai-compatible	https://openrouter.ai/api/v1	OPENROUTER_API_KEY
groq	openai-compatible	https://api.groq.com/openai/v1	GROQ_API_KEY
anthropic	anthropic-messages	https://api.anthropic.com	ANTHROPIC_API_KEY
mistral	openai-compatible	https://api.mistral.ai/v1	MISTRAL_API_KEY
together	openai-compatible	https://api.together.xyz/v1	TOGETHER_API_KEY

OpenAI-compatible providers (relay, custom, deepseek, openrouter, groq, mistral, together) share one client implementation. Anthropic uses a real Messages API adapter.

Configuring a Provider

Local Relay

[provider]
provider = "relay"
base_url = "http://127.0.0.1:1234/v1"
model = "Qwen3.6-35B-A3B-UD-IQ3_XXS.gguf"
context_window = 131072

DeepSeek

[provider]
provider = "deepseek"
model = "deepseek-chat"
api_key_env = "DEEPSEEK_API_KEY"
context_window = 64000
# Optional: override token pricing (defaults are preset)
input_price_per_1m_tokens = 0.27
output_price_per_1m_tokens = 1.10

OpenRouter

[provider]
provider = "openrouter"
model = "deepseek/deepseek-chat"
api_key_env = "OPENROUTER_API_KEY"
context_window = 64000

[provider.headers]
HTTP-Referer = "https://github.com/skaft/hamr"
X-Title = "Hamr"

OpenRouter automatically adds HTTP-Referer and X-Title default headers. You can override or add custom headers in the [provider.headers] section.

Groq

[provider]
provider = "groq"
model = "llama-3.3-70b-versatile"
api_key_env = "GROQ_API_KEY"
context_window = 128000

Anthropic (real Messages adapter)

[provider]
provider = "anthropic"
model = "frontier-sonnet-4-5-20250929"
api_key_env = "ANTHROPIC_API_KEY"
context_window = 200000

Anthropic uses the real Messages API (POST /v1/messages) with x-api-key auth, not the OpenAI-compatible format. System prompts map to the top-level system field. Tool use is supported for Anthropic via the Messages API native tool format.

Custom OpenAI-compatible

[provider]
provider = "custom"
base_url = "http://127.0.0.1:1234/v1"
model = "local-model"
api_key = "dummy"
context_window = 131072

Legacy Relay (backward compatible)

Your existing Relay/local config keeps working:

[provider]
kind = "openai-compatible"
base_url = "http://127.0.0.1:1234/v1"
model = "Qwen3.6-35B-A3B-UD-IQ3_XXS.gguf"

API Key Configuration

Prefer api_key_env over api_key. The environment variable is never written to disk.

api_key_env = "DEEPSEEK_API_KEY"  # reads from process.env.DEEPSEEK_API_KEY

Only use api_key for local servers that don't need auth:

api_key = "sk-no-key-required"  # local relay servers

Token Pricing & Session Spend

Cloud providers have preset token pricing. You can override pricing in config:

[provider]
provider = "deepseek"
input_price_per_1m_tokens = 0.27   # USD per 1M input tokens
output_price_per_1m_tokens = 1.10  # USD per 1M output tokens

Default pricing per provider:

Provider	Input ($/1M)	Output ($/1M)
DeepSeek	$0.27	$1.10
Groq	$0.59	$0.79
Anthropic	$3.00	$15.00
OpenRouter	varies	varies

When pricing is configured, the TUI shows a session spend indicator at the bottom of the screen: Spend: $0.004. Local providers show Spend: local.

Custom Headers

[provider.headers]
"HTTP-Referer" = "https://github.com/skaft/hamr"
"X-Title" = "Hamr"
"X-Custom" = "value"

Header values can reference environment variables using either $VAR or ${VAR} syntax:

[providers.relay.headers]
"CF-Access-Client-Id" = "${CF_ACCESS_CLIENT_ID}"
"CF-Access-Client-Secret" = "${CF_ACCESS_CLIENT_SECRET}"

Hamr resolves the variable from process.env at runtime. If the env var is unset or empty, the header is omitted entirely.

Multi-provider headers

When using the multi-provider config format, headers live under the provider definition:

[providers.relay]
enabled = true
compatibility = "openai-compatible"
base_url = "https://ai.watchyourtemper.com/v1"

[providers.relay.headers]
"CF-Access-Client-Id" = "${CF_ACCESS_CLIENT_ID}"
"CF-Access-Client-Secret" = "${CF_ACCESS_CLIENT_SECRET}"

Model Configuration

Each provider supports context window configuration:

[provider]
provider = "deepseek"
model = "deepseek-chat"
context_window = 64000

Live Provider Smoke Tests

Smoke tests verify end-to-end connectivity with real providers. They require user-supplied API keys and are not run in CI.

# Test Relay (local)
HAMR_LIVE_PROVIDER=relay bun run smoke:provider

# Test DeepSeek
HAMR_LIVE_PROVIDER=deepseek bun run smoke:provider

# Test OpenRouter
HAMR_LIVE_PROVIDER=openrouter bun run smoke:provider

# Test Groq
HAMR_LIVE_PROVIDER=groq bun run smoke:provider

# Test Anthropic
HAMR_LIVE_PROVIDER=anthropic bun run smoke:provider

# Test custom endpoint
HAMR_LIVE_PROVIDER=custom HAMR_CUSTOM_BASE_URL=http://127.0.0.1:1234/v1 bun run smoke:provider

Each smoke test sends a tiny prompt ("Reply with exactly: hamr-ok") and verifies the response. Streaming is also tested. Tests are skipped when required API keys are missing.

Switching Providers

From the TUI:

1. Press / and type settings
2. Navigate to the Model tab
3. Select Active Provider and choose from enabled providers
4. Select Active Model to pick a model from that provider
5. Changes persist automatically

Known Limitations

• Anthropic tool use is not yet supported. Requesting tools with an Anthropic provider will throw a clear error.
• Streaming is implemented in the shared OpenAI-compatible client but the Anthropic adapter only supports non-streaming requests.
• Mistral and Together are available as presets on the shared OpenAI-compatible client but lack dedicated smoke tests.