Serve the agent to a terminal client

[onechat] / turn.go

diff --git a/turn.go b/turn.go
index 76f524eefc5fff41012cb391fc144850d9cede2a..f8faad83d0a8be375b4e9fd3b2f9e6766284d2cf 100644 (file)
--- a/turn.go
+++ b/turn.go
@@ -5,6 +5,7 @@ import (
        "encoding/json"
        "errors"
        "fmt"
        "encoding/json"
        "errors"
        "fmt"

+       "path/filepath"

        "strings"
        "sync"
        "time"
        "strings"
        "sync"
        "time"


@@ -16,8 +17,8 @@ type modelClient interface {
 }
 
 // Event tells the clients what the agent does as it happens. Kind is the kind
 }
 
 // Event tells the clients what the agent does as it happens. Kind is the kind

-// of a logged message (user, reply, tool, echo), or one of busy, idle, and
-// error. ID is the ID of a logged message.
+// of a logged message (user, reply, tool, echo), or one of ready, busy, idle,
+// and error. ID is the ID of a logged message.

 type Event struct {
        Kind string `json:"kind"`
        ID   int    `json:"id,omitempty"`
 type Event struct {
        Kind string `json:"kind"`
        ID   int    `json:"id,omitempty"`


@@ -30,6 +31,9 @@ type Event struct {
 // input of turnInput. The loop runs the tool calls of each response and sends
 // their results back, until a response has no tool call: the final answer.
 //
 // input of turnInput. The loop runs the tool calls of each response and sends
 // their results back, until a response has no tool call: the final answer.
 //

+// 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.
+//

 // Each later request of a turn begins with the whole input of the request
 // before it, then its output items, then the tool results. The output items
 // hold the reasoning with its encrypted content, because the calls do not
 // Each later request of a turn begins with the whole input of the request
 // before it, then its output items, then the tool results. The output items
 // hold the reasoning with its encrypted content, because the calls do not


@@ -42,6 +46,7 @@ type loop struct {
        memory *Memory
        client modelClient
        tools  *toolbox
        memory *Memory
        client modelClient
        tools  *toolbox

+       layout *viewLayout

        clock  func() time.Time
        notify func(Event)
        wake   chan struct{}
        clock  func() time.Time
        notify func(Event)
        wake   chan struct{}


@@ -68,11 +73,17 @@ type functionCallOutput struct {
 
 // newLoop returns the loop of agent with its memory. Its calls go through
 // client, and notify gets each event. The clock gives the date of a turn.
 
 // newLoop returns the loop of agent with its memory. Its calls go through
 // client, and notify gets each event. The clock gives the date of a turn.

-func newLoop(agent *Agent, memory *Memory, client modelClient, clock func() time.Time, notify func(Event)) *loop {
+func newLoop(agent *Agent, memory *Memory, client modelClient, clock func() time.Time, notify func(Event)) (*loop, error) {
+       layout, err := loadLayout(filepath.Join(agent.dir, layoutFileName))
+       if err != nil {
+               return nil, err
+       }
+

        l := &loop{
                agent:  agent,
                memory: memory,
                client: client,
        l := &loop{
                agent:  agent,
                memory: memory,
                client: client,

+               layout: layout,

                clock:  clock,
                notify: notify,
                wake:   make(chan struct{}, 1),
                clock:  clock,
                notify: notify,
                wake:   make(chan struct{}, 1),


@@ -84,7 +95,7 @@ func newLoop(agent *Agent, memory *Memory, client modelClient, clock func() time
                workdir: agent.config.Workdir,
        }
 
                workdir: agent.config.Workdir,
        }
 

-       return l
+       return l, nil

 }
 
 // Submit logs a user message and wakes the loop. The message joins the next
 }
 
 // Submit logs a user message and wakes the loop. The message joins the next


@@ -122,6 +133,14 @@ func (l *loop) Interrupt() bool {
        return true
 }
 
        return true
 }
 

+// Busy tells if a turn runs.
+func (l *loop) Busy() bool {
+       l.mutex.Lock()
+       defer l.mutex.Unlock()
+
+       return l.cancel != nil
+}
+

 // Append logs a message and tells the clients. The toolbox logs its calls and
 // results through it.
 func (l *loop) Append(kind, text string) (Message, error) {
 // Append logs a message and tells the clients. The toolbox logs its calls and
 // results through it.
 func (l *loop) Append(kind, text string) (Message, error) {


@@ -246,14 +265,20 @@ func (l *loop) chain(ctx context.Context) error {
                skip[message.ID] = true
        }
 
                skip[message.ID] = true
        }
 

-       input := turnInput(l.memory.ViewWithout(skip), l.state(), messages)
-       for {
+       view, err := l.layout.Split(l.memory.ViewLines(skip))
+       if err != nil {
+               return err
+       }
+
+       input := turnInput(view, l.state(), messages)
+       for turn := "first"; ; turn = "next" {

                response, err := l.client.Create(ctx, Request{
                        Model:        l.agent.config.Model,
                        Effort:       l.agent.config.Effort,
                        Instructions: instructions,
                        Tools:        agentTools(),
                        Input:        input,
                response, err := l.client.Create(ctx, Request{
                        Model:        l.agent.config.Model,
                        Effort:       l.agent.config.Effort,
                        Instructions: instructions,
                        Tools:        agentTools(),
                        Input:        input,

+                       Turn:         turn,

                })
                if err != nil {
                        return err
                })
                if err != nil {
                        return err


@@ -332,21 +357,28 @@ func (l *loop) handle(ctx context.Context, input []any, response Response) ([]an
        return input, nil
 }
 
        return input, nil
 }
 

-// state returns the per-turn state: the date and time of the turn.
+// state returns the per-turn state: the end of the view, and the date and time
+// of the turn.

 func (l *loop) state() string {
 func (l *loop) state() string {

-       return "Now: " + l.clock().Format("Monday 2006-01-02 15:04 -07:00")
+       return "</chat>\n\nNow: " + l.clock().Format("Monday 2006-01-02 15:04 -07:00")

 }
 
 }
 

-// turnInput returns the input of the first request of a turn: one user message
-// with the view, the per-turn state, and each new message whole. The view
-// leaves out the leaf lines of the new messages, which come whole after it.
-func turnInput(view, state string, messages []Message) []any {
-       parts := []string{view + "\n", state + "\n"}
+// turnInput returns the input of the first request of a turn, stable parts
+// first: a user message for each view message, one with the per-turn state,
+// and one for each new message, whole. The view leaves out the leaf lines of
+// the new messages.
+func turnInput(view []string, state string, messages []Message) []any {
+       input := make([]any, 0, len(view)+1+len(messages))
+       for _, text := range view {
+               input = append(input, userMessage(text))
+       }
+
+       input = append(input, userMessage(state))

        for _, message := range messages {
        for _, message := range messages {

-               parts = append(parts, message.Text)
+               input = append(input, userMessage(message.Text))

        }
 
        }
 

-       return []any{userMessage(parts...)}
+       return input

 }
 
 // inputItem returns an output item as an input item for the next request. The
 }
 
 // inputItem returns an output item as an input item for the next request. The