X-Git-Url: https://git.giacomolli.uk/onechat/blobdiff_plain/5a93b6124068c2a8748fc0bfbb2a2d151815bcf0..03ef9525ba0e62037b55606b851ee746967a003f:/turn.go?ds=inline

diff --git a/turn.go b/turn.go
index e0bd489..4652321 100644
--- a/turn.go
+++ b/turn.go
@@ -10,6 +10,56 @@ import (
 	"time"
 )
 
+// Event tells the clients what the agent does as it happens. Kind is the kind
+// of a logged message (user, reply, tool, echo, note, work), or one of ready,
+// busy, idle, error, and subagent, which tells that the subagent Text starts a
+// run. ID is the ID of a logged message. Source names the client that sent a
+// user message, such as terminal or telegram, so that each client can show the
+// messages of the others, or the subagent of a report. Status tells how the run
+// of a report ended: finished, stopped, or failed. A private notify receipt
+// uses ID for its queue sequence and Status for accepted or duplicate, not
+// processed.
+type Event struct {
+	Kind   string `json:"kind"`
+	ID     int    `json:"id,omitempty"`
+	Text   string `json:"text,omitempty"`
+	Source string `json:"source,omitempty"`
+	Status string `json:"status,omitempty"`
+}
+
+// TurnLoop runs the turns of one agent. A user message goes to the log when it
+// comes, and it wakes the loop. A turn waits until every message before it is
+// summarized. Then the turn runs a Chain whose first request holds the prompt,
+// the tools, and the input of turnInput.
+//
+// The view goes out as the messages of a ViewLayout, so the first request of a
+// turn reads the view of the turn before from the prompt cache. User messages
+// that come during a turn join the next request of its chain after the tool
+// results, and so do the reports of the subagents. The subagents run until the
+// loop ends. Parent and child messages stay queued until the turn finishes.
+// active belongs to the running turn. The mutex protects pending, cancel, and
+// failure.
+type TurnLoop struct {
+	agent     *Agent
+	memory    *Memory
+	client    ChainClient
+	tools     *Toolbox
+	runs      *SubagentRuns
+	layout    *ViewLayout
+	clock     func() time.Time
+	logger    *slog.Logger
+	notify    func(Event)
+	wake      chan struct{}
+	messages  *DaemonMessages
+	openChild func(context.Context, *Agent) (*DaemonAgent, error)
+	active    []Message
+
+	mutex   sync.Mutex
+	pending []Message
+	cancel  context.CancelFunc
+	failure error
+}
+
 // newTurnLoop returns the loop of agent with its memory, its subagents, and the
 // WhatsApp link and the schedules of its tools. Its calls go through client,
 // and notify gets each event. The clock gives the date of a turn. logger gets
@@ -60,79 +110,25 @@ func newTurnLoop(agent *Agent, memory *Memory, subagents *Subagents, whatsapp *W
 		schedules: schedules,
 		workdir:   agent.config.Workdir,
 		agentName: agent.config.Name,
+		images:    agent.Images(),
 	}
 
-	return l, nil
-}
-
-// Event tells the clients what the agent does as it happens. Kind is the kind
-// of a logged message (user, reply, tool, echo, note, work), or one of ready,
-// busy, idle, error, and subagent, which tells that the subagent Text starts a
-// run. ID is the ID of a logged message. Source names the client that sent a
-// user message, such as terminal or telegram, so that each client can show the
-// messages of the others, or the subagent of a report. Status tells how the run
-// of a report ended: finished, stopped, or failed.
-type Event struct {
-	Kind   string `json:"kind"`
-	ID     int    `json:"id,omitempty"`
-	Text   string `json:"text,omitempty"`
-	Source string `json:"source,omitempty"`
-	Status string `json:"status,omitempty"`
-}
-
-// TurnLoop runs the turns of one agent. A user message goes to the log when it
-// comes, and it wakes the loop. A turn waits until every message before it is
-// summarized. Then the turn runs a Chain whose first request holds the prompt,
-// the tools, and the input of turnInput.
-//
-// The view goes out as the messages of a ViewLayout, so the first request of a
-// turn reads the view of the turn before from the prompt cache. User messages
-// that come during a turn join the next request of its chain after the tool
-// results, and so do the reports of the subagents. The subagents run until the
-// loop ends. Parent and child messages stay queued until the turn finishes.
-// active belongs to the running turn. The mutex protects pending, cancel, and
-// failure.
-type TurnLoop struct {
-	agent     *Agent
-	memory    *Memory
-	client    ChainClient
-	tools     *Toolbox
-	runs      *SubagentRuns
-	layout    *ViewLayout
-	clock     func() time.Time
-	logger    *slog.Logger
-	notify    func(Event)
-	wake      chan struct{}
-	messages  *DaemonMessages
-	openChild func(context.Context, *Agent) (*DaemonAgent, error)
-	active    []Message
-
-	mutex   sync.Mutex
-	pending []Message
-	cancel  context.CancelFunc
-	failure error
-}
-
-// Submit logs a user message from the client source and wakes the loop. The
-// message joins the next request: the first one of a new turn, or the next one
-// of a running turn.
-func (l *TurnLoop) Submit(source, text string) (Message, error) {
-	message, err := l.record("user", source, text)
-	if err != nil {
-		return Message{}, err
+	if agent.config.Parent == "" {
+		l.tools.imageShow = &ImageShow{store: agent.Images()}
 	}
 
-	l.mutex.Lock()
-
-	l.pending = append(l.pending, message)
-	l.mutex.Unlock()
+	return l, nil
+}
 
-	select {
-	case l.wake <- struct{}{}:
-	default:
+// state returns the per-turn state: the end of the view, and the date and time
+// of the turn, then the names and responsibilities of its children.
+func (l *TurnLoop) state() string {
+	state := "</chat>\n\nNow: " + l.clock().Format("Monday 2006-01-02 15:04 -07:00")
+	if l.messages != nil {
+		state += l.messages.ChildState(l.agent.config.Name)
 	}
 
-	return message, nil
+	return state
 }
 
 // Report logs the report text of the subagent name as a work message and tells
@@ -150,6 +146,26 @@ func (l *TurnLoop) Report(name, text, status string) {
 	}
 }
 
+// note logs a note and tells the clients. A note starts no turn.
+func (l *TurnLoop) note(text string) error {
+	_, err := l.record("note", "", text)
+
+	return err
+}
+
+// fail stops the loop with err.
+func (l *TurnLoop) fail(err error) {
+	l.mutex.Lock()
+
+	l.failure = errors.Join(l.failure, err)
+	l.mutex.Unlock()
+
+	select {
+	case l.wake <- struct{}{}:
+	default:
+	}
+}
+
 // Work logs text as a work message of source, such as the name of a subagent,
 // and tells the clients, with status. The message starts no turn until it goes
 // to Wake.
@@ -189,6 +205,46 @@ func (l *TurnLoop) Wake(messages ...Message) {
 	}
 }
 
+// record logs a message and tells the clients, with the client source of a user
+// message.
+func (l *TurnLoop) record(kind, source, text string) (Message, error) {
+	message, err := l.memory.Append(kind, text)
+	if err != nil {
+		return Message{}, err
+	}
+
+	l.notify(Event{
+		Kind:   kind,
+		ID:     message.ID,
+		Text:   text,
+		Source: source,
+	})
+
+	return message, nil
+}
+
+// Submit logs a user message from the client source and wakes the loop. The
+// message joins the next request: the first one of a new turn, or the next one
+// of a running turn.
+func (l *TurnLoop) Submit(source, text string) (Message, error) {
+	message, err := l.record("user", source, text)
+	if err != nil {
+		return Message{}, err
+	}
+
+	l.mutex.Lock()
+
+	l.pending = append(l.pending, message)
+	l.mutex.Unlock()
+
+	select {
+	case l.wake <- struct{}{}:
+	default:
+	}
+
+	return message, nil
+}
+
 // StopSubagent ends the run of the subagent name and tells if it ran. Its
 // report is logged, but it starts no turn.
 func (l *TurnLoop) StopSubagent(name string) (bool, error) {
@@ -229,10 +285,11 @@ func (l *TurnLoop) Flush(ctx context.Context) error {
 	return l.memory.Flush(ctx)
 }
 
-// Run runs a turn each time user messages or reports wait, until ctx ends. It
-// returns the error of a failed log write, which stops the agent. At its start
-// it resumes the subagents that a stop of the daemon cut, and at its end it
-// ends the runs of the subagents and waits for them.
+// Run starts a turn when user messages, reports, or queued external events
+// wait, until ctx ends. It returns the error of a failed log write, which stops
+// the agent. External events enter memory only between chains. At its start it
+// resumes the subagents that a stop of the daemon cut, and at its end it ends
+// the runs of the subagents and waits for them.
 func (l *TurnLoop) Run(ctx context.Context) error {
 	defer l.runs.Close()
 
@@ -242,12 +299,23 @@ func (l *TurnLoop) Run(ctx context.Context) error {
 	}
 
 	for {
-		for ctx.Err() == nil && l.waiting() {
+		for ctx.Err() == nil {
 			err := l.failed()
 			if err != nil {
 				return fmt.Errorf("check turn loop failure: %w", err)
 			}
 
+			if l.messages != nil {
+				err = l.messages.Notifications(l)
+				if err != nil {
+					return fmt.Errorf("prepare external events: %w", err)
+				}
+			}
+
+			if !l.waiting() {
+				break
+			}
+
 			err = l.turn(ctx)
 			if err != nil {
 				return fmt.Errorf("run turn: %w", err)
@@ -267,31 +335,6 @@ func (l *TurnLoop) Run(ctx context.Context) error {
 	}
 }
 
-// note logs a note and tells the clients. A note starts no turn.
-func (l *TurnLoop) note(text string) error {
-	_, err := l.record("note", "", text)
-
-	return err
-}
-
-// record logs a message and tells the clients, with the client source of a user
-// message.
-func (l *TurnLoop) record(kind, source, text string) (Message, error) {
-	message, err := l.memory.Append(kind, text)
-	if err != nil {
-		return Message{}, err
-	}
-
-	l.notify(Event{
-		Kind:   kind,
-		ID:     message.ID,
-		Text:   text,
-		Source: source,
-	})
-
-	return message, nil
-}
-
 // waiting tells if user messages or reports wait for a request.
 func (l *TurnLoop) waiting() bool {
 	l.mutex.Lock()
@@ -300,19 +343,6 @@ func (l *TurnLoop) waiting() bool {
 	return len(l.pending) > 0
 }
 
-// fail stops the loop with err.
-func (l *TurnLoop) fail(err error) {
-	l.mutex.Lock()
-
-	l.failure = errors.Join(l.failure, err)
-	l.mutex.Unlock()
-
-	select {
-	case l.wake <- struct{}{}:
-	default:
-	}
-}
-
 // failed returns the error of a failed write of a report.
 func (l *TurnLoop) failed() error {
 	l.mutex.Lock()
@@ -423,10 +453,19 @@ func (l *TurnLoop) chain(ctx context.Context) (string, error) {
 			Instructions: instructions,
 			Tools:        l.agent.Tools(),
 		},
-		take: l.take,
+		take:   l.take,
+		images: l.agent.Images(),
 	}
 
-	reply, err := chain.Run(ctx, turnInput(view, l.state(), messages))
+	input := turnInput(view, l.state(), messages)
+	for n, message := range messages {
+		input[len(input)-len(messages)+n], err = imageUserMessage(l.agent.Images(), message.Text)
+		if err != nil {
+			return "", fmt.Errorf("read turn images: %w", err)
+		}
+	}
+
+	reply, err := chain.Run(ctx, input)
 	if err != nil {
 		return "", fmt.Errorf("run turn chain: %w", err)
 	}
@@ -448,17 +487,6 @@ func (l *TurnLoop) take() []Message {
 	return messages
 }
 
-// state returns the per-turn state: the end of the view, and the date and time
-// of the turn, then the names and responsibilities of its children.
-func (l *TurnLoop) state() string {
-	state := "</chat>\n\nNow: " + l.clock().Format("Monday 2006-01-02 15:04 -07:00")
-	if l.messages != nil {
-		state += l.messages.ChildState(l.agent.config.Name)
-	}
-
-	return state
-}
-
 // turnError returns the text of the error that ended a turn.
 func turnError(ctx context.Context, err error) string {
 	if ctx.Err() != nil {