Documentation
¶
Overview ¶
Package llmprovider constructs cogito LLM transports from nib configuration.
Native adapters (anthropic, google) self-register via llmprovider/registry in their init(). This file registers the OpenAI-compatible factory and provides the top-level New / NewWithStore constructors.
Adding a new native protocol: write the adapter package, call registry.Register in init(), and blank-import it below. No changes to NewWithTemperatureAndStore are needed.
Index ¶
- Constants
- Variables
- func IsCodex(config types.ModelProviderConfig) bool
- func ListModelChoices(ctx context.Context, config types.ModelProviderConfig, store *auth.Store) (ids []string, partial bool, err error)
- func ListModels(ctx context.Context, config types.ModelProviderConfig, store *auth.Store) ([]string, error)
- func ModelsEndpoint(config types.ModelProviderConfig, store *auth.Store) (baseURL, apiKey string, err error)
- func New(config types.ModelProviderConfig) (cogito.LLM, error)
- func NewWithStore(config types.ModelProviderConfig, store *auth.Store) (cogito.LLM, error)
- func NewWithTemperature(config types.ModelProviderConfig, temperature float32) (cogito.LLM, error)
- func NewWithTemperatureAndStore(config types.ModelProviderConfig, temperature float32, store *auth.Store) (cogito.LLM, error)
- type ModelLister
Constants ¶
const ( ProviderOpenAI = "openai" ProviderCodex = "codex" )
Variables ¶
var ErrNoModelList = registry.ErrNoModelList
ErrNoModelList reports a provider whose protocol has no model listing nib can query. A UI should let the user type a model name instead.
Functions ¶
func IsCodex ¶
func IsCodex(config types.ModelProviderConfig) bool
func ListModelChoices ¶ added in v0.9.1
func ListModelChoices(ctx context.Context, config types.ModelProviderConfig, store *auth.Store) (ids []string, partial bool, err error)
ListModelChoices is ListModels plus whether the list is partial: a suggestion (see registry.PartialModelList) that a name outside it may still be valid for, so a UI should accept typed names instead of refusing them.
func ListModels ¶ added in v0.9.1
func ListModels(ctx context.Context, config types.ModelProviderConfig, store *auth.Store) ([]string, error)
ListModels returns the model IDs config's provider serves. OpenAI-compatible endpoints (and Ollama's /v1) are queried at /models; native protocols are asked through their adapter, with its own auth. Anything else returns ErrNoModelList.
func ModelsEndpoint ¶ added in v0.9.1
func ModelsEndpoint(config types.ModelProviderConfig, store *auth.Store) (baseURL, apiKey string, err error)
ModelsEndpoint returns the OpenAI-compatible base URL (the one /models hangs off) and key for config, resolving credentials the same way the LLM client does. Protocols without such an endpoint return ErrNoModelList.
func New ¶
func New(config types.ModelProviderConfig) (cogito.LLM, error)
New returns an independent LLM transport. OpenAI means any OpenAI-compatible HTTP endpoint, including a local LocalAI server.
func NewWithStore ¶ added in v0.9.0
NewWithStore is like New but resolves credentials from store for providers that support /login (anthropic, google, openai-direct).
func NewWithTemperature ¶
NewWithTemperature applies agent-specific sampling to OpenAI-compatible providers. Codex app-server owns its reasoning configuration and ignores the compatibility hint.
func NewWithTemperatureAndStore ¶ added in v0.9.0
func NewWithTemperatureAndStore(config types.ModelProviderConfig, temperature float32, store *auth.Store) (cogito.LLM, error)
NewWithTemperatureAndStore is the full constructor: temperature for OpenAI-compatible providers, credential store for providers with /login. A nil store is valid — the resolver falls through to config/env.
Codex is the only special case (it uses a subprocess, not HTTP). Every other provider goes through the registry: the provider Definition is looked up, and the registered adapter Factory for its Protocol is called.
Types ¶
Source Files
Directories
¶
| Path | Synopsis |
|---|---|
|
Package anthropic adapts the Anthropic Messages API to cogito.LLM.
|
Package anthropic adapts the Anthropic Messages API to cogito.LLM. |
|
Package azureresponses adapts the Azure OpenAI Responses API to cogito.LLM.
|
Package azureresponses adapts the Azure OpenAI Responses API to cogito.LLM. |
|
Package bedrock adapts the Amazon Bedrock Converse Stream API to cogito.LLM.
|
Package bedrock adapts the Amazon Bedrock Converse Stream API to cogito.LLM. |
|
Package catalog queries OpenAI-compatible API endpoints to discover per-model output token limits at runtime.
|
Package catalog queries OpenAI-compatible API endpoints to discover per-model output token limits at runtime. |
|
Package codex adapts the OpenAI Codex Responses API (ChatGPT backend) to cogito.LLM.
|
Package codex adapts the OpenAI Codex Responses API (ChatGPT backend) to cogito.LLM. |
|
Package copilot implements the GitHub Copilot multi-protocol adapter.
|
Package copilot implements the GitHub Copilot multi-protocol adapter. |
|
Package geminicli adapts the Google Cloud Code Assist API (used by Gemini CLI) to cogito.LLM.
|
Package geminicli adapts the Google Cloud Code Assist API (used by Gemini CLI) to cogito.LLM. |
|
Package google adapts the Google Generative AI API to cogito.LLM.
|
Package google adapts the Google Generative AI API to cogito.LLM. |
|
Package googlevertex adapts the Google Vertex AI API to cogito.LLM.
|
Package googlevertex adapts the Google Vertex AI API to cogito.LLM. |
|
Package ollama adapts the Ollama native chat API (/api/chat) to cogito.LLM.
|
Package ollama adapts the Ollama native chat API (/api/chat) to cogito.LLM. |
|
Package openairesponses adapts the OpenAI Responses API to cogito.LLM.
|
Package openairesponses adapts the OpenAI Responses API to cogito.LLM. |
|
Package registry maps wire protocols to adapter factories.
|
Package registry maps wire protocols to adapter factories. |