Files
CyberStrikeAI/internal/reasoning/provider.go
T
Sycun cfaa95d7bd refactor(runtime): remove Eino and make Pi the sole agent engine
Delete the Eino model adapters, callbacks, middleware, orchestration,
summarization and single-agent runner together with their dependencies.
Chat, batch tasks, workflow Agent nodes and sub-agents now share the Pi
runtime, and dispatch is a single code path instead of an engine switch.

Configuration migrates eino_middleware/eino_skills to runtime/skills and
drops the retired callback settings while old files still load. Settings,
i18n and docs describe Pi only. Also carry the Pi follow-up fixes: empty
required lists are omitted from tool schemas, historical tool results stay
with their call batch, model proxies work over HTTP/1.1, single-query
retrieval skips query expansion, and finalization docs match Pi lifecycle
events.
2026-11-12 17:55:36 +08:00

192 lines
4.7 KiB
Go

// Package reasoning resolves provider-neutral reasoning intent and request extensions.
package reasoning
import (
"strings"
"cyberstrike-ai/internal/config"
)
// ClientIntent is optional per-request override from ChatRequest.reasoning.
type ClientIntent struct {
Mode string
Effort string
}
// ThinkingLevel is the SDK-level selection, independent of provider wire fields.
func ThinkingLevel(oa *config.OpenAIConfig, client *ClientIntent) string {
if oa == nil {
return "off"
}
mode := effectiveModeForModel(oa, client)
if mode == "off" {
return "off"
}
effort := effectiveEffort(&oa.Reasoning, client, oa.Reasoning.AllowClientReasoningEffective())
if effort != "" {
return effort
}
if mode == "on" {
return "medium"
}
return "off"
}
type wireProfile int
const (
wireNone wireProfile = iota
wireClaude
wireDeepseek
wireOpenAI
wireOutputConfig
)
// ProviderExtraFields preserves channel-specific request options for Pi provider calls.
func ProviderExtraFields(oa *config.OpenAIConfig, client *ClientIntent) map[string]any {
if oa == nil {
return nil
}
sr := &oa.Reasoning
allowClient := sr.AllowClientReasoningEffective()
mode := effectiveModeForModel(oa, client)
fields := cloneExtraRequestFields(sr.ExtraRequestFields)
if mode == "off" {
clearReasoningExtraFields(fields)
if resolveWireProfile(oa, sr) == wireDeepseek {
if fields == nil {
fields = make(map[string]any)
}
fields["thinking"] = map[string]any{"type": "disabled"}
}
return openAIReasoningDisabledFields(fields, oa)
}
if strings.EqualFold(strings.TrimSpace(oa.Provider), "claude") ||
strings.EqualFold(strings.TrimSpace(oa.Provider), "anthropic") {
return fields
}
effort := effectiveEffort(sr, client, allowClient)
switch resolveWireProfile(oa, sr) {
case wireDeepseek:
if mode == "auto" || mode == "on" {
if fields == nil {
fields = make(map[string]any)
}
fields["thinking"] = map[string]any{"type": "enabled"}
}
if effort != "" {
if fields == nil {
fields = make(map[string]any)
}
fields["reasoning_effort"] = effortStringForAPI(effort)
}
case wireOutputConfig:
e := effort
if mode == "on" && e == "" {
e = "high"
}
if e != "" {
if fields == nil {
fields = make(map[string]any)
}
fields["output_config"] = map[string]any{"effort": effortStringForAPI(e)}
}
default:
e := effort
if mode == "on" && e == "" {
e = "medium"
}
if e != "" {
if fields == nil {
fields = make(map[string]any)
}
fields["reasoning_effort"] = effortStringForAPI(e)
}
}
return fields
}
func cloneExtraRequestFields(fields map[string]interface{}) map[string]any {
if len(fields) == 0 {
return nil
}
out := make(map[string]any, len(fields))
for k, v := range fields {
out[k] = v
}
return out
}
func clearReasoningExtraFields(fields map[string]any) {
for _, key := range []string{"thinking", "reasoning_effort", "output_config", "reasoning"} {
delete(fields, key)
}
}
// claudeSonnet37DefaultBudgetTokens matches Anthropic extended-thinking documentation examples (budget_tokens with max_tokens 16000).
func effectiveMode(sr *config.OpenAIReasoningConfig, client *ClientIntent, allowClient bool) string {
server := strings.ToLower(strings.TrimSpace(sr.ModeEffective()))
if server == "" || server == "default" {
server = "auto"
}
if !allowClient || client == nil {
return server
}
cm := strings.ToLower(strings.TrimSpace(client.Mode))
if cm == "" || cm == "default" {
return server
}
return cm
}
func effectiveEffort(sr *config.OpenAIReasoningConfig, client *ClientIntent, allowClient bool) string {
se := normalizeEffort(sr.Effort)
if !allowClient || client == nil {
return se
}
ce := normalizeEffort(client.Effort)
if ce != "" {
return ce
}
return se
}
func normalizeEffort(s string) string {
e := strings.ToLower(strings.TrimSpace(s))
switch e {
case "low", "medium", "high", "max", "xhigh":
return e
default:
return ""
}
}
func resolveWireProfile(oa *config.OpenAIConfig, sr *config.OpenAIReasoningConfig) wireProfile {
provider := strings.TrimSpace(oa.Provider)
if strings.EqualFold(provider, "claude") || strings.EqualFold(provider, "anthropic") {
return wireClaude
}
p := strings.ToLower(strings.TrimSpace(sr.ProfileEffective()))
switch p {
case "output_config", "output_config_effort":
return wireOutputConfig
case "openai", "openai_compat":
return wireOpenAI
case "deepseek", "deepseek_compat":
return wireDeepseek
case "auto", "":
if oa.IsDeepSeekEndpointOrModel() {
return wireDeepseek
}
return wireOpenAI
default:
return wireOpenAI
}
}
func effortStringForAPI(e string) string {
// 原样透传:OpenAI 官方多为 xhigh,部分兼容网关为 max,由配置/对话 effort 选择。
return strings.ToLower(strings.TrimSpace(e))
}