mirror of
https://github.com/coder/coder.git
synced 2026-09-24 15:04:27 +08:00
feat: add configurable reasoning effort to Coder agents (#26974)
This commit is contained in:
Danielle Maywood
+101
-58
@@ -10,6 +10,7 @@ import (
|
||||
"mime"
|
||||
"net/http"
|
||||
"net/url"
|
||||
"slices"
|
||||
"strconv"
|
||||
"strings"
|
||||
"time"
|
||||
@@ -107,26 +108,27 @@ const (
|
||||
|
||||
// Chat represents a chat session with an AI agent.
|
||||
type Chat struct {
|
||||
ID uuid.UUID `json:"id" format:"uuid"`
|
||||
OrganizationID uuid.UUID `json:"organization_id" format:"uuid"`
|
||||
OwnerID uuid.UUID `json:"owner_id" format:"uuid"`
|
||||
OwnerUsername string `json:"owner_username,omitempty"`
|
||||
OwnerName string `json:"owner_name,omitempty"`
|
||||
WorkspaceID *uuid.UUID `json:"workspace_id,omitempty" format:"uuid"`
|
||||
BuildID *uuid.UUID `json:"build_id,omitempty" format:"uuid"`
|
||||
AgentID *uuid.UUID `json:"agent_id,omitempty" format:"uuid"`
|
||||
ParentChatID *uuid.UUID `json:"parent_chat_id,omitempty" format:"uuid"`
|
||||
RootChatID *uuid.UUID `json:"root_chat_id,omitempty" format:"uuid"`
|
||||
LastModelConfigID uuid.UUID `json:"last_model_config_id" format:"uuid"`
|
||||
Title string `json:"title"`
|
||||
Status ChatStatus `json:"status"`
|
||||
PlanMode ChatPlanMode `json:"plan_mode,omitempty"`
|
||||
LastError *ChatError `json:"last_error,omitempty"`
|
||||
LastTurnSummary *string `json:"last_turn_summary"`
|
||||
DiffStatus *ChatDiffStatus `json:"diff_status,omitempty"`
|
||||
CreatedAt time.Time `json:"created_at" format:"date-time"`
|
||||
UpdatedAt time.Time `json:"updated_at" format:"date-time"`
|
||||
Archived bool `json:"archived"`
|
||||
ID uuid.UUID `json:"id" format:"uuid"`
|
||||
OrganizationID uuid.UUID `json:"organization_id" format:"uuid"`
|
||||
OwnerID uuid.UUID `json:"owner_id" format:"uuid"`
|
||||
OwnerUsername string `json:"owner_username,omitempty"`
|
||||
OwnerName string `json:"owner_name,omitempty"`
|
||||
WorkspaceID *uuid.UUID `json:"workspace_id,omitempty" format:"uuid"`
|
||||
BuildID *uuid.UUID `json:"build_id,omitempty" format:"uuid"`
|
||||
AgentID *uuid.UUID `json:"agent_id,omitempty" format:"uuid"`
|
||||
ParentChatID *uuid.UUID `json:"parent_chat_id,omitempty" format:"uuid"`
|
||||
RootChatID *uuid.UUID `json:"root_chat_id,omitempty" format:"uuid"`
|
||||
LastModelConfigID uuid.UUID `json:"last_model_config_id" format:"uuid"`
|
||||
LastReasoningEffort *string `json:"last_reasoning_effort,omitempty"`
|
||||
Title string `json:"title"`
|
||||
Status ChatStatus `json:"status"`
|
||||
PlanMode ChatPlanMode `json:"plan_mode,omitempty"`
|
||||
LastError *ChatError `json:"last_error,omitempty"`
|
||||
LastTurnSummary *string `json:"last_turn_summary"`
|
||||
DiffStatus *ChatDiffStatus `json:"diff_status,omitempty"`
|
||||
CreatedAt time.Time `json:"created_at" format:"date-time"`
|
||||
UpdatedAt time.Time `json:"updated_at" format:"date-time"`
|
||||
Archived bool `json:"archived"`
|
||||
// Shared is true when this chat's root chat has explicit user or group ACL entries.
|
||||
Shared bool `json:"shared"`
|
||||
PinOrder int32 `json:"pin_order"`
|
||||
@@ -548,13 +550,14 @@ type ToolResult struct {
|
||||
|
||||
// CreateChatRequest is the request to create a new chat.
|
||||
type CreateChatRequest struct {
|
||||
OrganizationID uuid.UUID `json:"organization_id" format:"uuid"`
|
||||
Content []ChatInputPart `json:"content"`
|
||||
SystemPrompt string `json:"system_prompt,omitempty"`
|
||||
WorkspaceID *uuid.UUID `json:"workspace_id,omitempty" format:"uuid"`
|
||||
ModelConfigID *uuid.UUID `json:"model_config_id,omitempty" format:"uuid"`
|
||||
MCPServerIDs []uuid.UUID `json:"mcp_server_ids,omitempty" format:"uuid"`
|
||||
Labels map[string]string `json:"labels,omitempty"`
|
||||
OrganizationID uuid.UUID `json:"organization_id" format:"uuid"`
|
||||
Content []ChatInputPart `json:"content"`
|
||||
SystemPrompt string `json:"system_prompt,omitempty"`
|
||||
WorkspaceID *uuid.UUID `json:"workspace_id,omitempty" format:"uuid"`
|
||||
ModelConfigID *uuid.UUID `json:"model_config_id,omitempty" format:"uuid"`
|
||||
ReasoningEffort *string `json:"reasoning_effort,omitempty"`
|
||||
MCPServerIDs []uuid.UUID `json:"mcp_server_ids,omitempty" format:"uuid"`
|
||||
Labels map[string]string `json:"labels,omitempty"`
|
||||
// UnsafeDynamicTools declares client-executed tools that the
|
||||
// LLM can invoke. This API is highly experimental and highly
|
||||
// subject to change.
|
||||
@@ -615,7 +618,8 @@ type CreateChatMessageRequest struct {
|
||||
BusyBehavior ChatBusyBehavior `json:"busy_behavior,omitempty" enums:"queue,interrupt"`
|
||||
// PlanMode switches the chat's persistent plan mode.
|
||||
// nil: no change, ptr to "plan": enable, ptr to "": clear.
|
||||
PlanMode *ChatPlanMode `json:"plan_mode,omitempty"`
|
||||
PlanMode *ChatPlanMode `json:"plan_mode,omitempty"`
|
||||
ReasoningEffort *string `json:"reasoning_effort,omitempty"`
|
||||
}
|
||||
|
||||
// EditChatMessageRequest is the request to edit a user message in a chat.
|
||||
@@ -624,7 +628,8 @@ type EditChatMessageRequest struct {
|
||||
// ModelConfigID, when set, overrides the model used for the
|
||||
// replacement user message and the assistant turn that follows.
|
||||
// When nil the original message's model is preserved.
|
||||
ModelConfigID *uuid.UUID `json:"model_config_id,omitempty" format:"uuid"`
|
||||
ModelConfigID *uuid.UUID `json:"model_config_id,omitempty" format:"uuid"`
|
||||
ReasoningEffort *string `json:"reasoning_effort,omitempty"`
|
||||
}
|
||||
|
||||
// CreateChatMessageResponse is the response from adding a message to a chat.
|
||||
@@ -775,15 +780,17 @@ func AllChatModelOverrideContexts() []ChatModelOverrideContext {
|
||||
// ChatModelOverrideResponse is the response body for the chat model override
|
||||
// configuration endpoint.
|
||||
type ChatModelOverrideResponse struct {
|
||||
Context ChatModelOverrideContext `json:"context"`
|
||||
ModelConfigID string `json:"model_config_id"`
|
||||
IsMalformed bool `json:"is_malformed"`
|
||||
Context ChatModelOverrideContext `json:"context"`
|
||||
ModelConfigID string `json:"model_config_id"`
|
||||
ReasoningEffort *string `json:"reasoning_effort,omitempty"`
|
||||
IsMalformed bool `json:"is_malformed"`
|
||||
}
|
||||
|
||||
// UpdateChatModelOverrideRequest is the request body for updating the chat
|
||||
// model override configuration endpoint.
|
||||
type UpdateChatModelOverrideRequest struct {
|
||||
ModelConfigID string `json:"model_config_id"`
|
||||
ModelConfigID string `json:"model_config_id"`
|
||||
ReasoningEffort *string `json:"reasoning_effort,omitempty"`
|
||||
}
|
||||
|
||||
// ChatPersonalModelOverrideContext identifies which chat context the user
|
||||
@@ -808,11 +815,12 @@ const (
|
||||
|
||||
// ChatPersonalModelOverride is a resolved user personal model override.
|
||||
type ChatPersonalModelOverride struct {
|
||||
Context ChatPersonalModelOverrideContext `json:"context"`
|
||||
Mode ChatPersonalModelOverrideMode `json:"mode"`
|
||||
ModelConfigID string `json:"model_config_id"`
|
||||
IsSet bool `json:"is_set"`
|
||||
IsMalformed bool `json:"is_malformed"`
|
||||
Context ChatPersonalModelOverrideContext `json:"context"`
|
||||
Mode ChatPersonalModelOverrideMode `json:"mode"`
|
||||
ModelConfigID string `json:"model_config_id"`
|
||||
ReasoningEffort *string `json:"reasoning_effort,omitempty"`
|
||||
IsSet bool `json:"is_set"`
|
||||
IsMalformed bool `json:"is_malformed"`
|
||||
}
|
||||
|
||||
// ChatPersonalModelOverrideDeploymentDefaults describes the deployment-level
|
||||
@@ -835,8 +843,9 @@ type UserChatPersonalModelOverridesResponse struct {
|
||||
// UpdateUserChatPersonalModelOverrideRequest is the request body for updating
|
||||
// a user personal model override.
|
||||
type UpdateUserChatPersonalModelOverrideRequest struct {
|
||||
Mode ChatPersonalModelOverrideMode `json:"mode"`
|
||||
ModelConfigID string `json:"model_config_id"`
|
||||
Mode ChatPersonalModelOverrideMode `json:"mode"`
|
||||
ModelConfigID string `json:"model_config_id"`
|
||||
ReasoningEffort *string `json:"reasoning_effort,omitempty"`
|
||||
}
|
||||
|
||||
// ChatPersonalModelOverridesAdminSettings describes whether users may manage
|
||||
@@ -1262,8 +1271,11 @@ type ChatModelConfig struct {
|
||||
ContextLimit int64 `json:"context_limit"`
|
||||
CompressionThreshold int32 `json:"compression_threshold"`
|
||||
ModelConfig *ChatModelCallConfig `json:"model_config,omitempty"`
|
||||
CreatedAt time.Time `json:"created_at" format:"date-time"`
|
||||
UpdatedAt time.Time `json:"updated_at" format:"date-time"`
|
||||
// ReasoningEfforts lists selectable reasoning effort values through
|
||||
// the model's configured maximum.
|
||||
ReasoningEfforts []string `json:"reasoning_efforts,omitempty"`
|
||||
CreatedAt time.Time `json:"created_at" format:"date-time"`
|
||||
UpdatedAt time.Time `json:"updated_at" format:"date-time"`
|
||||
}
|
||||
|
||||
// ChatModelProviderOptions contains typed provider-specific options.
|
||||
@@ -1289,7 +1301,6 @@ type ChatModelOpenAIProviderOptions struct {
|
||||
MaxToolCalls *int64 `json:"max_tool_calls,omitempty" description:"Maximum number of tool calls per response"`
|
||||
ParallelToolCalls *bool `json:"parallel_tool_calls,omitempty" description:"Whether the model may make multiple tool calls in parallel"`
|
||||
User *string `json:"user,omitempty" description:"Unique identifier for the end user for abuse monitoring" hidden:"true"`
|
||||
ReasoningEffort *string `json:"reasoning_effort,omitempty" description:"Controls the level of reasoning effort" enum:"none,minimal,low,medium,high,xhigh"`
|
||||
ReasoningSummary *string `json:"reasoning_summary,omitempty" description:"Controls whether reasoning tokens are summarized in the response" enum:"auto,concise,detailed"`
|
||||
MaxCompletionTokens *int64 `json:"max_completion_tokens,omitempty" description:"Upper bound on tokens the model may generate"`
|
||||
TextVerbosity *string `json:"text_verbosity,omitempty" description:"Controls the verbosity of the text response" enum:"low,medium,high"`
|
||||
@@ -1315,7 +1326,6 @@ type ChatModelAnthropicThinkingOptions struct {
|
||||
type ChatModelAnthropicProviderOptions struct {
|
||||
SendReasoning *bool `json:"send_reasoning,omitempty" description:"Whether to include reasoning content in the response"`
|
||||
Thinking *ChatModelAnthropicThinkingOptions `json:"thinking,omitempty" description:"Configuration for extended thinking"`
|
||||
Effort *string `json:"effort,omitempty" label:"Reasoning Effort" description:"Controls the level of reasoning effort" enum:"low,medium,high,xhigh,max"`
|
||||
ThinkingDisplay *string `json:"thinking_display,omitempty" label:"Thinking Display" description:"Controls how Anthropic returns thinking content" enum:"summarized,omitted"`
|
||||
DisableParallelToolUse *bool `json:"disable_parallel_tool_use,omitempty" description:"Whether to disable parallel tool execution"`
|
||||
WebSearchEnabled *bool `json:"web_search_enabled,omitempty" description:"Enable Anthropic web search tool for grounding responses with real-time information"`
|
||||
@@ -1346,17 +1356,15 @@ type ChatModelGoogleProviderOptions struct {
|
||||
|
||||
// ChatModelOpenAICompatProviderOptions configures OpenAI-compatible behavior.
|
||||
type ChatModelOpenAICompatProviderOptions struct {
|
||||
User *string `json:"user,omitempty" description:"Unique identifier for the end user for abuse monitoring" hidden:"true"`
|
||||
ReasoningEffort *string `json:"reasoning_effort,omitempty" description:"Controls the level of reasoning effort" enum:"none,minimal,low,medium,high,xhigh"`
|
||||
User *string `json:"user,omitempty" description:"Unique identifier for the end user for abuse monitoring" hidden:"true"`
|
||||
}
|
||||
|
||||
// ChatModelReasoningOptions configures reasoning behavior for model
|
||||
// providers that support it.
|
||||
type ChatModelReasoningOptions struct {
|
||||
Enabled *bool `json:"enabled,omitempty" description:"Whether reasoning is enabled"`
|
||||
Exclude *bool `json:"exclude,omitempty" description:"Whether to exclude reasoning content from the response"`
|
||||
MaxTokens *int64 `json:"max_tokens,omitempty" description:"Maximum number of tokens for reasoning output"`
|
||||
Effort *string `json:"effort,omitempty" description:"Controls the level of reasoning effort" enum:"none,minimal,low,medium,high,xhigh"`
|
||||
Enabled *bool `json:"enabled,omitempty" description:"Whether reasoning is enabled"`
|
||||
Exclude *bool `json:"exclude,omitempty" description:"Whether to exclude reasoning content from the response"`
|
||||
MaxTokens *int64 `json:"max_tokens,omitempty" description:"Maximum number of tokens for reasoning output"`
|
||||
}
|
||||
|
||||
// ChatModelOpenRouterProvider configures OpenRouter routing preferences.
|
||||
@@ -1409,16 +1417,51 @@ type ModelCostConfig struct {
|
||||
CacheWritePricePerMillionTokens *decimal.Decimal `json:"cache_write_price_per_million_tokens,omitempty" description:"Cache write or cache creation token price in USD per 1M tokens"`
|
||||
}
|
||||
|
||||
// Reasoning effort levels, ordered low to high for clamping and comparison.
|
||||
const (
|
||||
ChatModelReasoningEffortNone = "none"
|
||||
ChatModelReasoningEffortMinimal = "minimal"
|
||||
ChatModelReasoningEffortLow = "low"
|
||||
ChatModelReasoningEffortMedium = "medium"
|
||||
ChatModelReasoningEffortHigh = "high"
|
||||
ChatModelReasoningEffortXHigh = "xhigh"
|
||||
ChatModelReasoningEffortMax = "max"
|
||||
)
|
||||
|
||||
var chatModelReasoningEffortValues = []string{
|
||||
ChatModelReasoningEffortNone,
|
||||
ChatModelReasoningEffortMinimal,
|
||||
ChatModelReasoningEffortLow,
|
||||
ChatModelReasoningEffortMedium,
|
||||
ChatModelReasoningEffortHigh,
|
||||
ChatModelReasoningEffortXHigh,
|
||||
ChatModelReasoningEffortMax,
|
||||
}
|
||||
|
||||
// ChatModelReasoningEffortValues returns the global reasoning effort scale.
|
||||
func ChatModelReasoningEffortValues() []string {
|
||||
return slices.Clone(chatModelReasoningEffortValues)
|
||||
}
|
||||
|
||||
// ChatModelReasoningEffortConfig configures per-model reasoning effort
|
||||
// bounds. When configured, Default and Max must both be provided before
|
||||
// storing.
|
||||
type ChatModelReasoningEffortConfig struct {
|
||||
Default *string `json:"default,omitempty" label:"Default Reasoning Effort" description:"Reasoning effort used when the user has not selected one" enum:"none,minimal,low,medium,high,xhigh,max"`
|
||||
Max *string `json:"max,omitempty" label:"Max Reasoning Effort" description:"Maximum reasoning effort the user may select" enum:"none,minimal,low,medium,high,xhigh,max"`
|
||||
}
|
||||
|
||||
// ChatModelCallConfig configures per-call model behavior defaults.
|
||||
type ChatModelCallConfig struct {
|
||||
MaxOutputTokens *int64 `json:"max_output_tokens,omitempty" description:"Upper bound on tokens the model may generate"`
|
||||
Temperature *float64 `json:"temperature,omitempty" description:"Sampling temperature between 0 and 2"`
|
||||
TopP *float64 `json:"top_p,omitempty" description:"Nucleus sampling probability cutoff"`
|
||||
TopK *int64 `json:"top_k,omitempty" description:"Number of highest-probability tokens to keep for sampling"`
|
||||
PresencePenalty *float64 `json:"presence_penalty,omitempty" description:"Penalty for tokens that have already appeared in the output"`
|
||||
FrequencyPenalty *float64 `json:"frequency_penalty,omitempty" description:"Penalty for tokens based on their frequency in the output"`
|
||||
Cost *ModelCostConfig `json:"cost,omitempty" description:"Optional pricing metadata for this model"`
|
||||
ProviderOptions *ChatModelProviderOptions `json:"provider_options,omitempty" description:"Provider-specific option overrides"`
|
||||
MaxOutputTokens *int64 `json:"max_output_tokens,omitempty" description:"Upper bound on tokens the model may generate"`
|
||||
Temperature *float64 `json:"temperature,omitempty" description:"Sampling temperature between 0 and 2"`
|
||||
TopP *float64 `json:"top_p,omitempty" description:"Nucleus sampling probability cutoff"`
|
||||
TopK *int64 `json:"top_k,omitempty" description:"Number of highest-probability tokens to keep for sampling"`
|
||||
PresencePenalty *float64 `json:"presence_penalty,omitempty" description:"Penalty for tokens that have already appeared in the output"`
|
||||
FrequencyPenalty *float64 `json:"frequency_penalty,omitempty" description:"Penalty for tokens based on their frequency in the output"`
|
||||
Cost *ModelCostConfig `json:"cost,omitempty" description:"Optional pricing metadata for this model"`
|
||||
ReasoningEffort *ChatModelReasoningEffortConfig `json:"reasoning_effort,omitempty" description:"Default and max reasoning effort for the model"`
|
||||
ProviderOptions *ChatModelProviderOptions `json:"provider_options,omitempty" description:"Provider-specific option overrides"`
|
||||
}
|
||||
|
||||
// UnmarshalJSON accepts both the current nested cost object and the previous
|
||||
|
||||
+12
-10
@@ -23,13 +23,11 @@ func TestChatModelProviderOptions_MarshalJSON_UsesPlainProviderPayload(t *testin
|
||||
t.Parallel()
|
||||
|
||||
sendReasoning := true
|
||||
effort := "high"
|
||||
thinkingDisplay := "summarized"
|
||||
|
||||
raw, err := json.Marshal(codersdk.ChatModelProviderOptions{
|
||||
Anthropic: &codersdk.ChatModelAnthropicProviderOptions{
|
||||
SendReasoning: &sendReasoning,
|
||||
Effort: &effort,
|
||||
ThinkingDisplay: &thinkingDisplay,
|
||||
},
|
||||
})
|
||||
@@ -37,7 +35,6 @@ func TestChatModelProviderOptions_MarshalJSON_UsesPlainProviderPayload(t *testin
|
||||
require.NotContains(t, string(raw), `"type":"anthropic.options"`)
|
||||
require.NotContains(t, string(raw), `"data":`)
|
||||
require.Contains(t, string(raw), `"send_reasoning":true`)
|
||||
require.Contains(t, string(raw), `"effort":"high"`)
|
||||
require.Contains(t, string(raw), `"thinking_display":"summarized"`)
|
||||
}
|
||||
|
||||
@@ -47,7 +44,6 @@ func TestChatModelProviderOptions_UnmarshalJSON_ParsesPlainProviderPayloads(t *t
|
||||
raw := []byte(`{
|
||||
"anthropic": {
|
||||
"send_reasoning": true,
|
||||
"effort": "high",
|
||||
"thinking_display": "summarized"
|
||||
}
|
||||
}`)
|
||||
@@ -58,12 +54,6 @@ func TestChatModelProviderOptions_UnmarshalJSON_ParsesPlainProviderPayloads(t *t
|
||||
require.NotNil(t, decoded.Anthropic)
|
||||
require.NotNil(t, decoded.Anthropic.SendReasoning)
|
||||
require.True(t, *decoded.Anthropic.SendReasoning)
|
||||
require.NotNil(t, decoded.Anthropic.Effort)
|
||||
require.Equal(
|
||||
t,
|
||||
"high",
|
||||
*decoded.Anthropic.Effort,
|
||||
)
|
||||
require.NotNil(t, decoded.Anthropic.ThinkingDisplay)
|
||||
require.Equal(t, "summarized", *decoded.Anthropic.ThinkingDisplay)
|
||||
}
|
||||
@@ -288,6 +278,18 @@ func TestChatMessagePart_StripInternal(t *testing.T) {
|
||||
})
|
||||
}
|
||||
|
||||
func TestChatModelReasoningEffortConfigEnumTags(t *testing.T) {
|
||||
t.Parallel()
|
||||
|
||||
want := strings.Join(codersdk.ChatModelReasoningEffortValues(), ",")
|
||||
typ := reflect.TypeOf(codersdk.ChatModelReasoningEffortConfig{})
|
||||
for _, fieldName := range []string{"Default", "Max"} {
|
||||
field, ok := typ.FieldByName(fieldName)
|
||||
require.True(t, ok)
|
||||
require.Equal(t, want, field.Tag.Get("enum"))
|
||||
}
|
||||
}
|
||||
|
||||
// TestChatMessagePartVariantTags validates the `variants` struct tags
|
||||
// on ChatMessagePart fields. Every field must either declare variant
|
||||
// membership or be explicitly excluded, and every known part type
|
||||
|
||||
Reference in New Issue
Block a user