Models and Providers

Agent YAML selects connic-managed inference or a BYOK provider and configures model behavior.

Model controls

model: anthropic/claude-sonnet-4-5
fallback_model: connic/glm-5.2
retry_options:
  attempts: 3
  attempt_timeout: 120
temperature: 0.7
reasoning_effort: high  # supported values depend on the selected model

• model selects the primary model and is required for LLM agents.
• fallback_model receives the request after the first failed primary attempt. With retry_options.attempts: 3, this allows one primary attempt and up to three fallback attempts. Timeouts and provider rejections follow this policy. An unsupported_file_type error skips further attempts on the same model and can still switch to the fallback. Managed and BYOK models can pair in either direction. A switch marks the run with context.fallback_model_used.
• A context-window error first triggers configured context compression and one extra request to the same model. This repair is allowed once per logical model call, outside the normal attempt count. If it fails, the normal retry or fallback policy continues. A fallback switch is immediate; retries on the fallback use the configured backoff or a valid provider Retry-After.
• Every attempt has a request timeout, defaulting to 120 seconds, capped by the run’s remaining time. Cancellation, an expired run deadline, or output already delivered by the current request stops retries and fallback. An interrupted stream is reported as incomplete. Earlier completed model calls and tools keep their results.
• temperature controls output randomness. Lower values are more deterministic.
• reasoning_effort uses model-specific values. Use auto for model-managed behavior; for connic/*, choose an override from the model catalog. Providers can report reasoning-token usage without returning displayable reasoning text. An explicit supported effort requests reasoning text. Captured reasoning appears in run traces when the provider returns it; some models do not expose summaries at any effort.
• reasoning_budget sets a raw token budget on models that accept one. Use reasoning_effort for models with named effort levels.

Supported providers

Select a matching prefix in the agent configuration. Only BYOK providers require credentials in Project Settings.

Custom OpenAI-compatible providers

Connect any OpenAI-compatible endpoint, including self-hosted models and inference proxies. In Project Settings > LLM Provider, select Add Custom Provider and configure:

• Model prefix: a unique lowercase prefix, such as ollama
• API base URL: the endpoint, such as https://my-llm.example.com/v1
• API key: optional; leave it empty if the endpoint does not require authentication

# Custom provider configured with prefix "ollama"
model: ollama/llama3

# Custom provider configured with prefix "vllm"
model: vllm/mistral-7b

connic routes requests for that prefix to the configured endpoint. The prefix must not collide with a built-in provider name; connic is reserved.

Example model identifiers

# Using a connic-managed model
model: connic/glm-5.2

# Using OpenAI with your own key
model: openai/gpt-5.2

# Using Anthropic
model: anthropic/claude-opus-4-7

# Using Google Gemini
model: gemini/gemini-2.5-pro

# Using Azure OpenAI (use your deployment name)
model: azure/my-gpt5-deployment

# Using OpenRouter (provider/model format)
model: openrouter/anthropic/claude-sonnet-4.5

# Using AWS Bedrock
model: bedrock/us.anthropic.claude-opus-4-7-v1:0

# Using Google Vertex AI
model: vertex_ai/gemini-2.5-pro

# Using a custom OpenAI-compatible provider
model: ollama/llama3

Provider documentation

• connic model catalog
• OpenAI Models
• Anthropic Models
• Google Gemini Models
• Azure OpenAI Models
• OpenRouter Models
• AWS Bedrock Models
• Google Vertex AI Models