provider

type Attribution struct {
	Workspace string
	ThreadID  string
	TaskID    string
}

type ChatRequest struct {
	Model           string                 `json:"model,omitempty"`
	Messages        []protocol.ChatMessage `json:"messages"`
	Tools           []ToolSpec             `json:"tools,omitempty"`
	Stream          bool                   `json:"stream,omitempty"`
	Temperature     float64                `json:"temperature,omitempty"`
	MaxTokens       int                    `json:"max_tokens,omitempty"`
	ReasoningEffort string                 `json:"reasoning_effort,omitempty"`
	// StreamOptions carries OpenAI streaming options; ChatStream sets
	// include_usage so the final chunk reports token usage. Omitted on non-stream
	// requests (the usage block is always present in a non-stream response).
	StreamOptions *StreamOptions `json:"stream_options,omitempty"`
}

type ChatResponse struct {
	Message protocol.ChatMessage `json:"message"`
	// Model is the provider-echoed served model ("" if the provider omits it; the
	// caller falls back to the requested model).
	Model string `json:"model,omitempty"`
	// Usage is the provider-reported token accounting (zero when unavailable).
	Usage Usage `json:"usage,omitempty"`
}

type DeltaFunc func(kind DeltaKind, text string) error

type DeltaKind int

type FailureKind string

type Guard func(base *url.URL, authHeader string) error

type OpenAICompatClient struct {
	// contains filtered or unexported fields
}

type OpenAICompatConfig struct {
	BaseURL    string
	AuthHeader string
	// Auth supplies dynamic credentials (for example a renewable OAuth profile).
	// It is mutually exclusive with AuthHeader. The host owns credential storage.
	Auth       llmclient.Authenticator
	ChatPath   string
	Format     WireFormat
	HTTPClient *http.Client
	// PrepareRequest applies an optional provider-specific request profile after
	// neutral protocol conversion and before encoding.
	PrepareRequest func(*llmprotocol.Request)
	// StreamFirstChunk bounds the wait for the FIRST streamed chunk (time-to-first-
	// token — long for big-context reasoning). StreamIdle bounds the gap between
	// SUBSEQUENT chunks (once tokens flow, gaps are short). Neither is a total-duration
	// cap. 0 → defaults (see NewOpenAICompatClient).
	StreamFirstChunk time.Duration
	StreamIdle       time.Duration
	// Guard, when set, validates BaseURL + AuthHeader (e.g. SandboxGuard).
	Guard Guard
	// PromptCaching, when true, marks the end of the stable system-prompt prefix
	// with a provider prompt-cache breakpoint on the native wire path (the sidecar
	// lowers it to an Anthropic cache_control:{type:"ephemeral"} block). Opt-in and
	// default false so it cannot regress existing behavior; a no-op on the openai
	// wire format (which drops the hint).
	PromptCaching bool
	// StreamUsage, when true, requests a trailing usage chunk on streamed calls
	// (stream_options.include_usage) and reads it instead of short-circuiting on
	// finish_reason. Opt-in and default false: it must only be enabled against
	// providers that send a terminal `[DONE]`/usage chunk, else the stream waits on
	// the idle-liveness timer. Non-stream calls always report usage regardless.
	StreamUsage bool
}

type PromptCachePlan struct {
	StablePrefixBytes   int
	StablePromptDigest  string
	DynamicSuffixDigest string
	ToolSchemaDigest    string
}

type ProviderError struct {
	Kind   FailureKind
	Status int // HTTP status, 0 for transport-level failures
	Body   string
	// contains filtered or unexported fields
}

type StreamOptions struct {
	IncludeUsage bool `json:"include_usage,omitempty"`
}

type ToolSpec struct {
	Name        string
	Description string
	Parameters  json.RawMessage // JSON-schema object; empty => {"type":"object"}
}

type Usage struct {
	PromptTokens             int `json:"prompt_tokens,omitempty"`
	CompletionTokens         int `json:"completion_tokens,omitempty"`
	TotalTokens              int `json:"total_tokens,omitempty"`
	CachedInputTokens        int `json:"cached_input_tokens,omitempty"`
	CacheCreationInputTokens int `json:"cache_creation_input_tokens,omitempty"`
}

type WireFormat string