Files
alan 06001697c8 Add per-chat thinking levels for Codex, Claude Code and OpenCode
Codex: model/list supportedReasoningEfforts per model, turn/start effort,
thread reasoningEffort as the current level. Claude Code: initialize
supportedEffortLevels, apply_flag_settings effortLevel at the next turn
start. OpenCode v2: model variants via the session model endpoint.
chat_effort sets a level; rows and chat_models carry the levels; Codex and
Claude levels persist in operator_sessions.chat_effort and are re-applied
before every turn.
2026-11-09 20:07:41 +00:00

135 lines
4.3 KiB
Go

package codex
import (
"context"
"encoding/json"
"errors"
"time"
"git.a-wan.net/alan/pipelines/internal/protocol"
)
// catalogTTL bounds how long a model/list read answers rows and choices; the
// account's catalog changes rarely, and a session list must not issue one
// model/list per row.
const catalogTTL = time.Minute
// catalogModel is one visible model/list entry. Model.model, not the catalog
// row's opaque id, is the turn/start override.
type catalogModel struct {
ID string
Name string
Efforts []protocol.ChatEffort
DefaultEffort string
}
// Models uses the thread's actual provider and the account's paginated catalog.
func (c *Chat) Models(ctx context.Context, id string) ([]protocol.ChatModel, error) {
session, found, err := c.Session(ctx, id)
if err != nil {
return nil, err
}
if !found || session.ModelProvider == "" {
return nil, errors.New("Codex thread model provider is unavailable")
}
catalog, err := c.catalog(ctx)
if err != nil {
return nil, err
}
models := make([]protocol.ChatModel, 0, len(catalog))
for _, model := range catalog {
models = append(models, protocol.ChatModel{ID: model.ID, Provider: session.ModelProvider, Name: model.Name, Efforts: model.Efforts, DefaultEffort: model.DefaultEffort})
}
return models, nil
}
// catalog returns the visible model/list entries, read at most once a minute.
func (c *Chat) catalog(ctx context.Context) ([]catalogModel, error) {
c.mu.Lock()
cached, at := c.models, c.modelsAt
c.mu.Unlock()
if cached != nil && time.Since(at) < catalogTTL {
return cached, nil
}
models, err := c.readCatalog(ctx)
if err != nil {
return nil, err
}
c.mu.Lock()
c.models, c.modelsAt = models, time.Now()
c.mu.Unlock()
return models, nil
}
func (c *Chat) readCatalog(ctx context.Context) ([]catalogModel, error) {
models := []catalogModel{}
cursor := ""
seen := map[string]bool{}
for page := 0; page < 20; page++ {
params := map[string]any{"limit": 100, "includeHidden": false}
if cursor != "" {
params["cursor"] = cursor
}
raw, err := c.call(ctx, "model/list", params)
if err != nil {
return nil, err
}
var response struct {
Data *[]struct {
Model string `json:"model"`
DisplayName string `json:"displayName"`
Hidden bool `json:"hidden"`
DefaultReasoningEffort string `json:"defaultReasoningEffort"`
SupportedReasoningEfforts []struct {
ReasoningEffort string `json:"reasoningEffort"`
Description string `json:"description"`
} `json:"supportedReasoningEfforts"`
} `json:"data"`
NextCursor *string `json:"nextCursor"`
}
if err := json.Unmarshal(raw, &response); err != nil || response.Data == nil {
return nil, errors.New("invalid Codex model catalog")
}
for _, model := range *response.Data {
if model.Model == "" || model.Hidden {
continue
}
entry := catalogModel{ID: model.Model, Name: model.DisplayName}
for _, option := range model.SupportedReasoningEfforts {
if option.ReasoningEffort != "" && !protocol.HasChatEffort(entry.Efforts, option.ReasoningEffort) {
entry.Efforts = append(entry.Efforts, protocol.ChatEffort{ID: option.ReasoningEffort, Name: protocol.ChatEffortName(option.ReasoningEffort), Description: option.Description})
}
}
if protocol.HasChatEffort(entry.Efforts, model.DefaultReasoningEffort) {
entry.DefaultEffort = model.DefaultReasoningEffort
}
models = append(models, entry)
}
if response.NextCursor == nil || *response.NextCursor == "" {
return models, nil
}
cursor = *response.NextCursor
if seen[cursor] {
return nil, errors.New("Codex model catalog repeated its cursor")
}
seen[cursor] = true
}
return nil, errors.New("Codex model catalog exceeds pagination limit")
}
// SwitchModel validates a selection only. Pipelines persists it before replying
// and supplies it to the next explicit turn/start. Never use thread/resume here:
// resumption may start a queued goal and is not a safe settings operation.
func (c *Chat) SwitchModel(ctx context.Context, id, provider, model string) error {
models, err := c.Models(ctx, id)
if err != nil {
return err
}
for _, available := range models {
if available.ID == model && available.Provider == provider {
return nil
}
}
return errors.New("the selected Codex model is not available for this thread's provider")
}