conversations

package
v2.5.1 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Jul 29, 2026 License: Apache-2.0 Imports: 26 Imported by: 0

Documentation

Index

Constants

View Source 
const (
	AgentMentionReminderBotUserIDProp      = "bot_user_id"
	AgentMentionReminderBotUsernameProp    = "bot_username"
	AgentMentionReminderBotDisplayNameProp = "bot_display_name"
	AgentMentionReminderTargetPostIDProp   = "target_post_id"
)

Prop keys carried on the custom post for the webapp to read when rendering.

View Source 
const (
	ActivateAIProp   = "activate_ai"
	FromWebhookProp  = "from_webhook"
	FromBotProp      = "from_bot"
	FromPluginProp   = "from_plugin"
	FromOAuthAppProp = "from_oauth_app"
	WranglerProp     = "wrangler"
)
View Source 
const (
	ReferencedRecordingFileID  = "referenced_recording_file_id"
	ReferencedTranscriptPostID = "referenced_transcript_post_id"
)
View Source 
const AgentMentionReminderPostType = "custom_agent_mention_reminder"

AgentMentionReminderPostType is the custom post type used for the ephemeral "you must @mention an agent" reminder rendered by the webapp.

View Source 
const AnalysisTypeProp = "prompt_type"
View Source 
const ThreadIDProp = "referenced_thread"

Variables

View Source 
var (
	ErrLoopInNotPostOwner       = errors.New("only the post author can loop in an agent")
	ErrLoopInNotThreadReply     = errors.New("loop in is only available for thread replies")
	ErrLoopInUnsupportedChannel = errors.New("loop in is not available in direct or group messages")
	ErrLoopInAlreadyMentioned   = errors.New("post already mentions an agent")
	ErrLoopInWrongAgent         = errors.New("previous post in thread was not authored by the requested agent")
	ErrLoopInNoAgentContext     = errors.New("no agent context for this thread reply")
)
View Source 
var ErrInvalidToolAnswer = errors.New("invalid answer for user interaction tool call")

ErrInvalidToolAnswer is returned when an accepted user-interaction tool call (e.g. AskUserQuestion) arrives without a valid answer. The pending state is left untouched so the user can answer again. The HTTP layer maps this to 400 Bad Request.

View Source 
var (
	// ErrNoResponse is returned when no response is posted under a normal condition.
	ErrNoResponse = errors.New("no response")
)
View Source 
var ErrNotRequester = errors.New("only the original requester can approve/reject tool calls")

ErrNotRequester is returned when a user other than the original conversation requester attempts to approve or reject tool calls/results. The HTTP layer maps this to 403 Forbidden.

View Source 
var ErrPostMissingConversationID = errors.New("post missing conversation_id")

ErrPostMissingConversationID is returned when a tool-approval request arrives for a post that has no conversation_id prop. The HTTP layer maps this to 400 Bad Request.

View Source 
var ErrRemoteMCPNotLicensed = errors.New("tools from remote MCP servers require a license with MCP support")

ErrRemoteMCPNotLicensed is returned when a tool decision would execute, or share the output of, a tool served by a remote/external MCP server on a server whose license does not include MCP support. Built-in tools (empty ServerOrigin) and embedded Mattermost MCP tools (mcp.EmbeddedClientKey) are basic tool integrations and never require a license. The HTTP layer maps this to 403 Forbidden.

This gate is the decision-time backstop for pending remote tool calls persisted before a license change; the primary enforcement is at supply time, where llmcontext.Builder drops remote MCP tools from the LLM context entirely on unlicensed servers.

View Source 
var ErrStaleToolClick = errors.New("stale or duplicate tool-approval click")

ErrStaleToolClick is returned when a tool-approval click cannot be resolved because the pending tool state no longer matches the request. Typical causes: another browser tab already approved/rejected, the post is not an approval post, or the approval has expired. The HTTP layer maps this to 400 Bad Request rather than 500 Internal Server Error.

Functions

This section is empty.

Types

type ConfigProvider 

type ConfigProvider interface {
	EnableChannelMentionToolCalling() bool
	AllowNativeWebSearchInChannels() bool
	MCP() mcp.Config
}

ConfigProvider provides configuration values for conversation behavior

type Conversations 

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

func New 

func New(
	prompts *llm.Prompts,
	mmClient mmapi.Client,
	streamingService streaming.Service,
	contextBuilder *llmcontext.Builder,
	botsService *bots.MMBots,
	db *mmapi.DBClient,
	licenseChecker *enterprise.LicenseChecker,
	i18nBundle *i18n.Bundle,
	meetingsService MeetingsService,
	configProvider ConfigProvider,
) *Conversations

func (*Conversations) CreateOrGetDMConversation 

func (c *Conversations) CreateOrGetDMConversation(
	botID string,
	postingUser *model.User,
	channel *model.Channel,
	post *model.Post,
	llmCtx *llm.Context,
) (*DMConversationResult, error)

CreateOrGetDMConversation creates or retrieves a conversation for a DM. This is separated from ProcessDMRequest so the conversation_id can be set on the response post before it is created.

func (*Conversations) DeleteConversationsForDeletedPost 

func (c *Conversations) DeleteConversationsForDeletedPost(post *model.Post) error

DeleteConversationsForDeletedPost soft-deletes conversations associated with the given post. If the post is a root post, conversations keyed by that RootPostID are marked as deleted.

func (*Conversations) HandleLoopInAgent 

func (c *Conversations) HandleLoopInAgent(ctx context.Context, userID string, bot *bots.Bot, post *model.Post, channel *model.Channel) error

HandleLoopInAgent processes the user's thread reply through the conversation flow when they click "loop in" on the agent mention reminder.

func (*Conversations) HandleRegenerate 

func (c *Conversations) HandleRegenerate(ctx stdcontext.Context, userID string, post *model.Post, channel *model.Channel) error

HandleRegenerate handles post regeneration requests

func (*Conversations) HandleToolCall 

func (c *Conversations) HandleToolCall(ctx context.Context, userID string, post *model.Post, channel *model.Channel, acceptedToolIDs []string, toolAnswers map[string]mmtools.UserInteractionAnswer) error

HandleToolCall handles user approval/rejection of pending tool calls via conversation entities. It looks up pending tool_use blocks in the conversation turns, executes approved tools, writes results back as turns, and streams a follow-up LLM response.

toolAnswers carries the user's answers for accepted user-interaction tool calls (e.g. AskUserQuestion), keyed by tool_use block ID. Those blocks are not executed; the validated answer becomes the tool result.

func (*Conversations) HandleToolResult 

func (c *Conversations) HandleToolResult(ctx context.Context, userID string, post *model.Post, channel *model.Channel, acceptedToolIDs []string) error

HandleToolResult handles user approval of the second-stage tool-result sharing. It flips shared flags for accepted results and, in channels, streams the LLM follow-up with unshared content redacted so private tool output cannot leak into the channel-visible reply.

func (*Conversations) MessageHasBeenPosted 

func (c *Conversations) MessageHasBeenPosted(_ *plugin.Context, post *model.Post)

func (*Conversations) ProcessDMRequest 

func (c *Conversations) ProcessDMRequest(
	ctx stdcontext.Context,
	convID string,
	lm llm.LanguageModel,
	llmCtx *llm.Context,
	maxToolTurns int,
) (*DMStreamResult, error)

ProcessDMRequest builds a completion request from the conversation and runs the tool loop, returning the final stream. The conversation must already exist (created via CreateOrGetDMConversation).

maxToolTurns bounds the tool-call-execute-recall loop for this bot; pass 0 or any non-positive value to use the system default (llm.DefaultMaxToolTurns).

func (*Conversations) SaveTitle 

func (c *Conversations) SaveTitle(threadID, title string) error

SaveTitle saves a title for a conversation identified by its root post ID. It looks up the conversation by RootPostID and updates the title via the LLM_Conversations table.

func (*Conversations) SaveTitleAsync 

func (c *Conversations) SaveTitleAsync(threadID, title string)

SaveTitleAsync saves a title asynchronously

func (*Conversations) SetConversationService 

func (c *Conversations) SetConversationService(svc *conversation.Service)

SetConversationService sets the conversation entity service.

func (*Conversations) SetMeetingsService 

func (c *Conversations) SetMeetingsService(meetingsService MeetingsService)

SetMeetingsService sets the meetings service (used to break circular dependency during initialization)

func (*Conversations) SetToolPolicyChecker 

func (c *Conversations) SetToolPolicyChecker(checker mcp.ToolPolicyChecker)

SetToolPolicyChecker sets the per-tool policy checker used for auto-approval and DM auto-run decisions.

type DMConversationResult 

type DMConversationResult struct {
	ConversationID string
	IsNew          bool
	UserTurnID     string
}

DMConversationResult is the return value of CreateOrGetDMConversation.

type DMStreamResult 

type DMStreamResult struct {
	Stream *llm.TextStreamResult
}

DMStreamResult is the return value of ProcessDMRequest.

type MeetingsService 

type MeetingsService interface {
	GetCaptionsFileIDFromProps(post *model.Post) (fileID string, err error)
	SummarizeTranscription(ctx stdcontext.Context, bot *bots.Bot, transcription *subtitles.Subtitles, context *llm.Context) (*llm.TextStreamResult, error)
}

MeetingsService defines the interface for meetings functionality needed by conversations

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.