llm

Documentation

Overview

Package llm is the canonical provider port contract for the hawk ecosystem.

It is the single source of truth for the conversation DTOs and the Provider interface that hawk (product face) and eyrie (provider engine) speak across their boundary. Both sides alias to these types, so there is exactly one definition of each DTO and no per-call conversion.

hawk owns the product vocabulary (hence names like EyrieMessage); eyrie implements the port. eyrie's internal transport types stay eyrie-scoped and never appear here.

Index

• type CacheControl
• type CatalogHealth
• type CatalogMaintenance
• type CatalogSnapshot
• type ChatOptions
• type CheckStatus
• type ContentPart
• type ContextManagement
• type ContinuationConfig
• type CredentialManager
• type CredentialProviderOption
• type CredentialResolution
• type CredentialStatus
• type DeploymentSummary
• type EventStreamer
• type EyrieConfig
• type EyrieMessage
• type EyrieResponse
• type EyrieStreamEvent
• type EyrieTool
• type EyrieUsage
• type Gateway
• type GatewayInspector
• type GatewayRegionOption
• type GenerateRequest
• type GenerationOptions
• type Generator
• type ImageURLPart
• type InputAudioPart
• type Intent
• type Limits
• type Metadata
• type Model
• type ModelCatalog
• type ModelClass
• type NativeCompactionRequest
• type NativeCompactor
• type Preference
• type PreflightCheck
• type PreflightOptions
• type PreflightReport
• type PromptCacheOpts
• type Provider
• type ProviderStateSecurity
• type Requirements
• type ResolvedRoute
• type ResponseFormat
• type RoutingConfig
• type Selection
• type SelectionManager
• type SelectionOptions
• type StatePaths
• type StreamResult
  • func NewStreamResult(events <-chan EyrieStreamEvent, requestID string, cancel context.CancelFunc) *StreamResult
  • func (sr *StreamResult) Close()
• type ToolCall
• type ToolChoiceOption
• type ToolResult

Types

type CacheControl

type CacheControl struct {
	Type string `json:"type"`
	TTL  string `json:"ttl,omitempty"`
}

type CatalogHealth

type CatalogHealth struct {
	Path        string    `json:"path"`
	Exists      bool      `json:"exists"`
	ModifiedAt  time.Time `json:"modified_at,omitempty"`
	Size        int64     `json:"size,omitempty"`
	Models      int       `json:"models,omitempty"`
	Deployments int       `json:"deployments,omitempty"`
	Offerings   int       `json:"offerings,omitempty"`
	Stale       bool      `json:"stale,omitempty"`
	StaleAfter  time.Time `json:"stale_after,omitempty"`
	Source      string    `json:"source,omitempty"`
	Error       string    `json:"error,omitempty"`
}

CatalogHealth reports the health of the model catalog (host-facing).

type CatalogMaintenance

type CatalogMaintenance interface {
	RefreshCatalog(ctx context.Context, providerID string) (CatalogSnapshot, error)
	CatalogHealth(ctx context.Context) CatalogHealth
	StatePaths() StatePaths
	DefaultProviderFilter(ctx context.Context) string
	Preflight(ctx context.Context) PreflightReport
	PreflightWithOptions(ctx context.Context, opts PreflightOptions) PreflightReport
	ProviderStateSecurityStatus() ProviderStateSecurity
	MigrateProviderSecrets() error
	MigrateProviderSecretsContext(ctx context.Context) error
}

CatalogMaintenance is the refresh/preflight/security facet (config only).

type CatalogSnapshot

type CatalogSnapshot struct {
	Models    []Model   `json:"models"`
	CachePath string    `json:"cache_path,omitempty"`
	RemoteURL string    `json:"remote_url,omitempty"`
	Stale     bool      `json:"stale,omitempty"`
	LoadedAt  time.Time `json:"loaded_at"`
}

CatalogSnapshot is an immutable, point-in-time host-facing view of a loaded model catalog. It is the canonical definition; eyrie's engine.CatalogSnapshot is a type alias to this so a single struct crosses the host boundary.

type ChatOptions

type ChatOptions struct {
	Provider             string          `json:"provider,omitempty"`
	Model                string          `json:"model,omitempty"`
	Temperature          *float64        `json:"temperature,omitempty"`
	MaxTokens            int             `json:"max_tokens,omitempty"`
	Stream               bool            `json:"stream,omitempty"`
	Tools                []EyrieTool     `json:"tools,omitempty"`
	System               string          `json:"system,omitempty"`
	EnableCaching        bool            `json:"enable_caching,omitempty"`
	ResponseFormat       *ResponseFormat `json:"response_format,omitempty"`
	ReasoningEffort      string          `json:"reasoning_effort,omitempty"`
	ThinkingBudgetTokens int             `json:"thinking_budget_tokens,omitempty"`
	ThinkingMode         string          `json:"thinking_mode,omitempty"`
	ThinkingDisplay      string          `json:"thinking_display,omitempty"`
	ThinkingEnabled      *bool           `json:"thinking_enabled,omitempty"`
	// GLMThinkingEnabled is a deprecated alias of ThinkingEnabled kept for
	// older Z.AI call sites. New code should set ThinkingEnabled. Adapters
	// accept either field; wire encoding is selected per provider via
	// OpenAICompatConfig.ThinkingFormat ("zai", "longcat", "agnes", …).
	GLMThinkingEnabled *bool             `json:"glm_thinking_enabled,omitempty"`
	VirtualKeyID       string            `json:"virtual_key_id,omitempty"`
	KimiContextCacheID string            `json:"kimi_context_cache_id,omitempty"`
	KimiCacheResetTTL  bool              `json:"kimi_cache_reset_ttl,omitempty"`
	TopP               *float64          `json:"top_p,omitempty"`
	TopK               *int              `json:"top_k,omitempty"`
	StopSequences      []string          `json:"stop_sequences,omitempty"`
	ToolChoice         *ToolChoiceOption `json:"tool_choice,omitempty"`
	MetadataUserID     string            `json:"metadata_user_id,omitempty"`
	ServiceTier        string            `json:"service_tier,omitempty"`
	OutputEffort       string            `json:"output_effort,omitempty"`
	OutputSchema       string            `json:"output_schema,omitempty"`
	PresencePenalty    *float64          `json:"presence_penalty,omitempty"`
	FrequencyPenalty   *float64          `json:"frequency_penalty,omitempty"`
	N                  *int              `json:"n,omitempty"`
	LogProbs           *bool             `json:"logprobs,omitempty"`
	TopLogProbs        *int              `json:"top_logprobs,omitempty"`
	Seed               *int              `json:"seed,omitempty"`
	Store              *bool             `json:"store,omitempty"`
	Metadata           map[string]string `json:"metadata,omitempty"`
	Modalities         []string          `json:"modalities,omitempty"`
	AudioConfig        string            `json:"audio_config,omitempty"`
	Prediction         string            `json:"prediction,omitempty"`
	WebSearchOptions   string            `json:"web_search_options,omitempty"`
	PromptCacheOptions *PromptCacheOpts  `json:"prompt_cache_options,omitempty"`
	CacheControl       *CacheControl     `json:"cache_control,omitempty"`
	// Responses API extensions (Concentrate)
	PreviousResponseID string              `json:"previous_response_id,omitempty"`
	Include            []string            `json:"include,omitempty"`
	Routing            *RoutingConfig      `json:"routing,omitempty"`
	MaxToolCalls       *int                `json:"max_tool_calls,omitempty"`
	ContextManagement  []ContextManagement `json:"context_management,omitempty"`
	SafetyIdentifier   *string             `json:"safety_identifier,omitempty"`
	Background         *bool               `json:"background,omitempty"`
	Truncation         *string             `json:"truncation,omitempty"`
	Conversation       interface{}         `json:"conversation,omitempty"`
}

ChatOptions holds request options for an engine chat call.

type CheckStatus

type CheckStatus string

CheckStatus is a preflight check status string.

const (
	CheckOK   CheckStatus = "ok"
	CheckFail CheckStatus = "fail"
	CheckWarn CheckStatus = "warn"
)

Common preflight check statuses.

type ContentPart

type ContentPart struct {
	Type       string          `json:"type"`
	Text       string          `json:"text,omitempty"`
	ImageURL   *ImageURLPart   `json:"image_url,omitempty"`
	InputAudio *InputAudioPart `json:"input_audio,omitempty"`
}

ContentPart is a provider-neutral multimodal message part.

type ContextManagement

type ContextManagement struct {
	Type string `json:"type"` // "compact_20260112", "clear_thinking_20251015", etc.
}

type ContinuationConfig

type ContinuationConfig struct {
	MaxContinuations int
	MaxTotalTokens   int
}

ContinuationConfig controls output continuation behavior.

type CredentialManager

type CredentialManager interface {
	SaveCredential(ctx context.Context, providerID, secret string) (CredentialStatus, error)
	RemoveCredential(ctx context.Context, providerID string) error
	CredentialStatus(ctx context.Context, providerID string) (CredentialStatus, error)
	SaveCredentialEnv(ctx context.Context, envVar, secret string) error
	HasCredentialEnv(ctx context.Context, envVar string) bool
	CredentialEnvKeys(providerID string) []string
	ResolveCredential(ctx context.Context, secret string) CredentialResolution
	CredentialProviders(context.Context) []CredentialProviderOption
	ApplyCredentials(ctx context.Context, providerID string) (CatalogSnapshot, error)
}

CredentialManager is the key/credential facet (config only).

type CredentialProviderOption

type CredentialProviderOption struct {
	ProviderID   string `json:"provider_id"`
	DeploymentID string `json:"deployment_id,omitempty"`
	EnvVar       string `json:"env_var"`
	DisplayName  string `json:"display_name"`
	Inferred     bool   `json:"inferred,omitempty"`
	RequiresKey  bool   `json:"requires_key"`
	Rank         int    `json:"rank"`
}

CredentialProviderOption is one provider row in the key picker.

type CredentialResolution

type CredentialResolution struct {
	FormatOK                bool   `json:"format_ok"`
	FormatError             string `json:"format_error,omitempty"`
	Providers               []CredentialProviderOption
	ProbeDisambiguationUsed bool `json:"probe_disambiguation_used,omitempty"`
}

CredentialResolution is the result of validating a pasted API key.

type CredentialStatus

type CredentialStatus struct {
	Configured          bool   `json:"configured"`
	ProviderID          string `json:"provider_id,omitempty"`
	EnvironmentVariable string `json:"environment_variable,omitempty"`
	EnvironmentConflict bool   `json:"environment_conflict,omitempty"`
	Verified            bool   `json:"verified,omitempty"`
	Masked              string `json:"masked,omitempty"`
	EnvVar              string `json:"env_var,omitempty"`
}

CredentialStatus reports whether a provider's credential is configured.

type DeploymentSummary

type DeploymentSummary struct {
	RoutingSource string `json:"routing_source,omitempty"`
	RoutingStages int    `json:"routing_stages,omitempty"`
	Formatted     string `json:"formatted"`
}

DeploymentSummary summarizes deployment routing for a model (host-facing).

type EventStreamer

type EventStreamer interface {
	Next() bool
	Event() EyrieStreamEvent
	Err() error
	Close() error
}

EventStreamer is the pull-based host stream contract used by the engine facade. Next must not be called concurrently. Close is idempotent.

type EyrieConfig

type EyrieConfig struct {
	Provider   string `json:"provider,omitempty"`
	APIKey     string `json:"-"`
	BaseURL    string `json:"base_url,omitempty"`
	Model      string `json:"model,omitempty"`
	MaxRetries int    `json:"max_retries,omitempty"`
}

EyrieConfig holds client configuration.

type EyrieMessage

type EyrieMessage struct {
	Role         string        `json:"role"`
	Content      string        `json:"content,omitempty"`
	Thinking     string        `json:"thinking,omitempty"`
	ContentParts []ContentPart `json:"content_parts,omitempty"`
	Images       []string      `json:"images,omitempty"`
	ToolUse      []ToolCall    `json:"tool_use,omitempty"`
	ToolResults  []ToolResult  `json:"tool_results,omitempty"`
}

EyrieMessage is the provider-neutral conversation message shape.

type EyrieResponse

type EyrieResponse struct {
	Content        string         `json:"content"`
	Thinking       string         `json:"thinking,omitempty"`
	Usage          *EyrieUsage    `json:"usage,omitempty"`
	ToolCalls      []ToolCall     `json:"tool_calls,omitempty"`
	FinishReason   string         `json:"finish_reason"`
	RequestID      string         `json:"request_id,omitempty"`
	OrganizationID string         `json:"organization_id,omitempty"`
	Route          *ResolvedRoute `json:"route,omitempty"`
}

EyrieResponse is the chat response DTO.

type EyrieStreamEvent

type EyrieStreamEvent struct {
	Type       string      `json:"type"`
	Content    string      `json:"content,omitempty"`
	ToolCall   *ToolCall   `json:"tool_call,omitempty"`
	Thinking   string      `json:"thinking,omitempty"`
	Error      string      `json:"error,omitempty"`
	Warning    string      `json:"warning,omitempty"`
	RequestID  string      `json:"request_id,omitempty"`
	Usage      *EyrieUsage `json:"usage,omitempty"`
	StopReason string      `json:"stop_reason,omitempty"`
	// TTFT and TTFTms both carry time-to-first-token in milliseconds but ride
	// different events: the dedicated "ttft" event populates TTFT, while the
	// terminal "done" event populates TTFTms. The engine normalizes the two
	// into a single value (preferring TTFTms, falling back to TTFT). Both are
	// retained for wire compatibility with existing producers/consumers.
	TTFTms int            `json:"ttft_ms,omitempty"`
	TTFT   int            `json:"ttft,omitempty"`
	Route  *ResolvedRoute `json:"route,omitempty"`
}

EyrieStreamEvent is a streaming event.

type EyrieTool

type EyrieTool struct {
	Name        string                 `json:"name"`
	Description string                 `json:"description"`
	Parameters  map[string]interface{} `json:"parameters"`
}

EyrieTool is a tool definition.

type EyrieUsage

type EyrieUsage struct {
	PromptTokens        int `json:"prompt_tokens"`
	CompletionTokens    int `json:"completion_tokens"`
	TotalTokens         int `json:"total_tokens"`
	CacheCreationTokens int `json:"cache_creation_tokens,omitempty"`
	CacheReadTokens     int `json:"cache_read_tokens,omitempty"`
	ThinkingTokens      int `json:"thinking_tokens,omitempty"`
}

EyrieUsage tracks token usage.

type Gateway

type Gateway struct {
	ID                    string                `json:"id"`
	DisplayName           string                `json:"display_name"`
	DeploymentID          string                `json:"deployment_id,omitempty"`
	CredentialEnv         string                `json:"credential_env"`
	RequiresKey           bool                  `json:"requires_key"`
	SortOrder             int                   `json:"sort_order"`
	ChatPreference        int                   `json:"chat_preference"`
	SupportsLiveDiscovery bool                  `json:"supports_live_discovery"`
	CredentialConfigured  bool                  `json:"credential_configured"`
	DeploymentConfigured  bool                  `json:"deployment_configured"`
	ModelCount            int                   `json:"model_count"`
	RegionLabel           string                `json:"region_label,omitempty"`
	RegionRequired        bool                  `json:"region_required"`
	RegionOptions         []GatewayRegionOption `json:"region_options,omitempty"`
	DNSHost               string                `json:"dns_host,omitempty"`
	Active                bool                  `json:"active"`
}

Gateway is a provider/gateway descriptor.

type GatewayInspector

type GatewayInspector interface {
	GatewayDefinitions() []Gateway
	Gateways(ctx context.Context) []Gateway
	GatewayRegion(providerID string) (label string, required bool)
	SetGatewayRegion(ctx context.Context, providerID, value string) error
	GatewayForModel(ctx context.Context, modelID string) string
	CanonicalModel(ctx context.Context, modelID string) string
	DeploymentRoutingEnabled(override *bool) bool
	DeploymentStatus(ctx context.Context, activeModel string) (string, error)
	DeploymentSummary(ctx context.Context, activeModel string) (DeploymentSummary, error)
	RoutingPreview(ctx context.Context, modelID string) (string, error)
}

GatewayInspector is the gateway/deployment facet (config only).

type GatewayRegionOption

type GatewayRegionOption struct {
	Value       string `json:"value"`
	DisplayName string `json:"display_name"`
	Endpoint    string `json:"endpoint,omitempty"`
}

GatewayRegionOption describes one selectable region for a regional gateway.

type GenerateRequest

type GenerateRequest struct {
	Messages     []EyrieMessage
	SystemPrompt string
	Tools        []EyrieTool
	Requirements Requirements
	Preference   Preference
	Limits       Limits
	Metadata     Metadata
	Temperature  *float64
	OutputSchema string
	Options      GenerationOptions
}

GenerateRequest is the normalized generation request.

type GenerationOptions

type GenerationOptions struct {
	EnableCaching        bool
	ReasoningEffort      string
	ThinkingBudgetTokens int
	ThinkingMode         string
	ThinkingDisplay      string
	ThinkingEnabled      *bool
	// GLMThinkingEnabled is a deprecated alias of ThinkingEnabled (Z.AI-era).
	GLMThinkingEnabled *bool
	VirtualKeyID       string
	KimiContextCacheID string
	KimiCacheResetTTL  bool
	TopP               *float64
	TopK               *int
	StopSequences      []string
	ToolChoice         *ToolChoiceOption
	ServiceTier        string
	OutputEffort       string
	PresencePenalty    *float64
	FrequencyPenalty   *float64
	N                  *int
	LogProbs           *bool
	TopLogProbs        *int
	Seed               *int
	Store              *bool
	Metadata           map[string]string
	Modalities         []string
	AudioConfig        string
	Prediction         string
	WebSearchOptions   string
}

GenerationOptions holds provider-specific generation knobs.

type Generator

type Generator interface {
	Generate(ctx context.Context, req GenerateRequest) (*EyrieResponse, error)
	Stream(ctx context.Context, req GenerateRequest) (EventStreamer, error)
}

Generator is the chat transport facet: the only part the ChatClient path uses.

Stream returns a pull-based EventStreamer (the host engine facade contract). Lower-level channel-based streaming still uses StreamResult on the client transport layer; that type is intentionally not part of this host port.

type ImageURLPart

type ImageURLPart struct {
	URL    string `json:"url"`
	Detail string `json:"detail,omitempty"`
}

ImageURLPart describes an image URL or data URI.

type InputAudioPart

type InputAudioPart struct {
	Data   string `json:"data"`
	Format string `json:"format"`
}

InputAudioPart describes base64-encoded audio content.

type Intent

type Intent string

Intent expresses a host's semantic preference without naming a provider.

const (
	IntentFast       Intent = "fast"
	IntentBalanced   Intent = "balanced"
	IntentReasoning  Intent = "reasoning"
	IntentEconomical Intent = "economical"
)

type Limits

type Limits struct {
	MaxOutputTokens      int           `json:"max_output_tokens,omitempty"`
	MaxContinuations     int           `json:"max_continuations,omitempty"`
	MaxTotalOutputTokens int           `json:"max_total_output_tokens,omitempty"`
	Timeout              time.Duration `json:"timeout,omitempty"`
}

Limits declares output limits.

type Metadata

type Metadata struct {
	SessionID string `json:"session_id,omitempty"`
	TurnID    string `json:"turn_id,omitempty"`
	UserID    string `json:"user_id,omitempty"`
	ProjectID string `json:"project_id,omitempty"`
}

Metadata carries request-scoped metadata.

type Model

type Model struct {
	ID               string          `json:"id"`
	ProviderID       string          `json:"provider_id"`
	CanonicalID      string          `json:"canonical_id,omitempty"`
	DisplayName      string          `json:"display_name"`
	Description      string          `json:"description,omitempty"`
	Owner            string          `json:"owner,omitempty"`
	GatewayID        string          `json:"gateway_id,omitempty"`
	ContextWindow    int             `json:"context_window,omitempty"`
	MaxOutputTokens  int             `json:"max_output_tokens,omitempty"`
	InputPricePer1M  float64         `json:"input_price_per_1m,omitempty"`
	OutputPricePer1M float64         `json:"output_price_per_1m,omitempty"`
	PriceKnown       bool            `json:"price_known"`
	Capabilities     []string        `json:"capabilities,omitempty"`
	Source           string          `json:"source,omitempty"`
	LiveMetadata     json.RawMessage `json:"live_metadata,omitempty"`

	// SupportsThinkingToggle indicates this model's provider wire protocol
	// honors the ThinkingEnabled toggle. Populated by the engine catalog.
	SupportsThinkingToggle bool `json:"supports_thinking_toggle,omitempty"`
	// DefaultThinkingEnabled is the provider-default thinking state for this
	// model when no explicit user preference is set. nil means no provider
	// default (fall through to global settings).
	DefaultThinkingEnabled *bool `json:"default_thinking_enabled,omitempty"`
}

Model is the product-facing view of model metadata.

type ModelCatalog

type ModelCatalog interface {
	ListModels(ctx context.Context, providerID string, refresh bool) ([]Model, error)
	ListLiveModels(ctx context.Context, providerID string) ([]Model, error)
	ListPublicModels(ctx context.Context, providerID string) ([]Model, error)
	ModelInfo(ctx context.Context, modelID string) (Model, bool, error)
	ModelProviders(ctx context.Context) ([]string, error)
	DefaultModel(ctx context.Context, provider, fallback string) string
	PreferredModel(ctx context.Context, provider string, class ModelClass, fallback string) string
	PreferredModels(ctx context.Context, primaryProvider string, class ModelClass, limit int) []string
	ModelClassOf(ctx context.Context, modelID string) ModelClass
	ProviderForModel(ctx context.Context, modelID string) string
	PrimaryModel(ctx context.Context) string
	ModelNames(ctx context.Context) []string
	Catalog(ctx context.Context) (CatalogSnapshot, error)
}

ModelCatalog is the model-discovery facet (used by routing + config).

type ModelClass

type ModelClass string

ModelClass is a provider-neutral relative model cost/capability band.

const (
	ModelClassEconomical ModelClass = "economical"
	ModelClassBalanced   ModelClass = "balanced"
	ModelClassPremium    ModelClass = "premium"
)

type NativeCompactionRequest

type NativeCompactionRequest struct {
	Provider        string
	Model           string
	Messages        []EyrieMessage
	ContextWindow   int
	ThresholdPct    int
	MaxOutputTokens int
}

NativeCompactionRequest is a native-compaction request.

type NativeCompactor

type NativeCompactor interface {
	SupportsNativeCompaction(ctx context.Context, provider, model string) bool
	CompactNative(ctx context.Context, req NativeCompactionRequest) (string, error)
}

NativeCompactor is the provider-native-compaction facet.

type Preference

type Preference struct {
	Intent            Intent  `json:"intent,omitempty"`
	PreferredProvider string  `json:"-"`
	PreferredModelID  string  `json:"preferred_model_id,omitempty"`
	AllowFallback     bool    `json:"allow_fallback,omitempty"`
	MaximumCostUSD    float64 `json:"maximum_cost_usd,omitempty"`
}

Preference declares the preferred provider/model.

type PreflightCheck

type PreflightCheck struct {
	Name   string      `json:"name"`
	Status CheckStatus `json:"status"`
	Detail string      `json:"detail,omitempty"`
}

PreflightCheck is one readiness check.

type PreflightOptions

type PreflightOptions struct {
	VerifyLive bool `json:"verify_live,omitempty"`
}

PreflightOptions configures a preflight check.

type PreflightReport

type PreflightReport struct {
	Ready        bool             `json:"ready"`
	LiveVerified bool             `json:"live_verified"`
	Checks       []PreflightCheck `json:"checks"`
}

PreflightReport is the result of a preflight check.

type PromptCacheOpts

type PromptCacheOpts struct {
	Mode string `json:"mode,omitempty"`
	TTL  string `json:"ttl,omitempty"`
}

type Provider

type Provider interface {
	Generator
	ModelCatalog
	CredentialManager
	SelectionManager
	GatewayInspector
	CatalogMaintenance
	NativeCompactor
}

Provider is hawk's hawk-owned view of the provider engine: a composition of the role interfaces below. It is the single integration surface — hawk never holds an *eyrieengine.Engine, and eyrie never imports hawk/internal.

Callers that need only a subset depend on the relevant role interface directly (e.g. session_factory depends only on Generator), keeping the declared dependency precise and the test stub small.

type ProviderStateSecurity

type ProviderStateSecurity struct {
	Path       string `json:"path"`
	HasSecrets bool   `json:"has_secrets"`
	Detail     string `json:"detail,omitempty"`
	Error      string `json:"error,omitempty"`
}

ProviderStateSecurity reports security state of provider config.

type Requirements

type Requirements struct {
	Streaming      bool
	Tools          bool
	Vision         bool
	StructuredJSON bool
	Reasoning      bool
	MinimumContext int `json:"minimum_context,omitempty"`
}

Requirements declare what the request needs from the engine.

type ResolvedRoute

type ResolvedRoute struct {
	Provider          string `json:"provider"`
	Model             string `json:"model"`
	DeploymentRouting bool   `json:"deployment_routing,omitempty"`
}

ResolvedRoute is the concrete provider/model route selected by the engine.

type ResponseFormat

type ResponseFormat struct {
	Type   string `json:"type"`
	Schema string `json:"schema,omitempty"`
}

ResponseFormat specifies the desired output format for a model response.

type RoutingConfig

type RoutingConfig struct {
	Model    interface{} `json:"model,omitempty"`
	Strategy string      `json:"strategy,omitempty"` // "min" or "max"
	Metric   string      `json:"metric,omitempty"`   // "cost", "latency", "quality"
}

type Selection

type Selection struct {
	Provider                string `json:"provider"`
	Model                   string `json:"model"`
	HasConfiguredDeployment bool   `json:"has_configured_deployment"`
	DeploymentRouting       bool   `json:"deployment_routing"`
}

Selection is the effective provider/model pair.

type SelectionManager

type SelectionManager interface {
	ActiveSelection(ctx context.Context) ResolvedRoute
	EffectiveSelection(ctx context.Context, opts SelectionOptions) Selection
	SetActiveProvider(ctx context.Context, provider string) error
	SetActiveModel(ctx context.Context, modelID string) error
	SetSelection(ctx context.Context, provider, modelID string) error
	ClearSelection(ctx context.Context) error
}

SelectionManager is the get/set selection facet (config only).

type SelectionOptions

type SelectionOptions struct {
	ProviderOverride          string
	ModelOverride             string
	DeploymentRoutingOverride *bool
}

SelectionOptions controls effective-selection resolution.

type StatePaths

type StatePaths struct {
	Catalog        string `json:"catalog"`
	ProviderConfig string `json:"provider_config"`
}

StatePaths is the on-disk location of catalog + provider config.

type StreamResult

type StreamResult struct {
	Events    <-chan EyrieStreamEvent
	RequestID string
	// contains filtered or unexported fields
}

StreamResult wraps a streaming response with cleanup. Callers must call Close() when done reading events, or cancel the context.

func NewStreamResult

func NewStreamResult(events <-chan EyrieStreamEvent, requestID string, cancel context.CancelFunc) *StreamResult

NewStreamResult constructs a stream result. The cancel function is optional and must be idempotent.

func (*StreamResult) Close

func (sr *StreamResult) Close()

Close stops the stream and releases resources.

type ToolCall

type ToolCall = tools.ToolCall

ToolCall is a tool invocation. Aliased to tools.ToolCall so the ecosystem has a single ToolCall/ToolResult vocabulary (see tools/tool.go).

type ToolChoiceOption

type ToolChoiceOption struct {
	Type                   string `json:"type"`
	Name                   string `json:"name,omitempty"`
	DisableParallelToolUse bool   `json:"disable_parallel_tool_use,omitempty"`
}

ToolChoiceOption controls how the model uses tools.

type ToolResult

type ToolResult = tools.ToolResult

ToolResult is a tool execution result. Aliased to tools.ToolResult.