X-Git-Url: https://git.giacomolli.uk/onechat/blobdiff_plain/5a93b6124068c2a8748fc0bfbb2a2d151815bcf0..80e348724041a694002fc38f57e8868b9b9a2be9:/turn.go

diff --git a/turn.go b/turn.go
index e0bd489..b5aa4cf 100644
--- a/turn.go
+++ b/turn.go
@@ -10,6 +10,54 @@ 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.
+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
@@ -60,57 +108,14 @@ 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
+	if agent.config.Parent == "" {
+		l.tools.imageShow = &ImageShow{store: agent.Images()}
+	}
 
-	mutex   sync.Mutex
-	pending []Message
-	cancel  context.CancelFunc
-	failure error
+	return l, nil
 }
 
 // Submit logs a user message from the client source and wakes the loop. The
@@ -423,10 +428,19 @@ 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(),
+	}
+
+	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, turnInput(view, l.state(), messages))
+	reply, err := chain.Run(ctx, input)
 	if err != nil {
 		return "", fmt.Errorf("run turn chain: %w", err)
 	}