chore: add sys.chat.current

go.mod

@@ -16,7 +16,7 @@ require (
 	github.com/gptscript-ai/broadcaster v0.0.0-20240625175512-c43682019b86
 	github.com/gptscript-ai/chat-completion-client v0.0.0-20240531200700-af8e7ecf0379
 	github.com/gptscript-ai/cmd v0.0.0-20240625175447-4250b42feb7d
-	github.com/gptscript-ai/tui v0.0.0-20240625175717-1e6eca7a66c1
+	github.com/gptscript-ai/tui v0.0.0-20240627001757-8b452fa47eb5
 	github.com/hexops/autogold/v2 v2.2.1
 	github.com/hexops/valast v1.4.4
 	github.com/jaytaylor/html2text v0.0.0-20230321000545-74c2419ad056
@@ -94,6 +94,7 @@ require (
 	github.com/pterm/pterm v0.12.79 // indirect
 	github.com/rivo/uniseg v0.4.7 // indirect
 	github.com/russross/blackfriday/v2 v2.1.0 // indirect
+	github.com/sourcegraph/go-diff-patch v0.0.0-20240223163233-798fd1e94a8e // indirect
 	github.com/ssor/bom v0.0.0-20170718123548-6386211fdfcf // indirect
 	github.com/therootcompany/xz v1.0.1 // indirect
 	github.com/tidwall/match v1.1.1 // indirect

go.sum

@@ -173,8 +173,8 @@ github.com/gptscript-ai/cmd v0.0.0-20240625175447-4250b42feb7d h1:sKf7T7twhGXs6A
 github.com/gptscript-ai/cmd v0.0.0-20240625175447-4250b42feb7d/go.mod h1:DJAo1xTht1LDkNYFNydVjTHd576TC7MlpsVRl3oloVw=
 github.com/gptscript-ai/go-gptscript v0.0.0-20240625134437-4b83849794cc h1:ABV7VAK65YBkqL7VlNp5ryVXnRqkKQ+U/NZfUO3ypqA=
 github.com/gptscript-ai/go-gptscript v0.0.0-20240625134437-4b83849794cc/go.mod h1:Dh6vYRAiVcyC3ElZIGzTvNF1FxtYwA07BHfSiFKQY7s=
-github.com/gptscript-ai/tui v0.0.0-20240625175717-1e6eca7a66c1 h1:sx/dJ0IRh3P9Ehr1g1TQ/jEw83KISmQyjrssVgPGUbE=
-github.com/gptscript-ai/tui v0.0.0-20240625175717-1e6eca7a66c1/go.mod h1:R33cfOnNaqsEn9es5jLKR39wvDyHvsIVgeTMNqtzCb8=
+github.com/gptscript-ai/tui v0.0.0-20240627001757-8b452fa47eb5 h1:knDhTTJNqaZB1XMudXJuVVnTqj9USrXzNfsl1nTqKXA=
+github.com/gptscript-ai/tui v0.0.0-20240627001757-8b452fa47eb5/go.mod h1:NwFdBDmGQvjLFFDnSRBRakkhw0MIO1sSdRnWNk4cCQ0=
 github.com/hashicorp/errwrap v1.0.0 h1:hLrqtEDnRye3+sgx6z4qVLNuviH3MR5aQ0ykNJa/UYA=
 github.com/hashicorp/errwrap v1.0.0/go.mod h1:YH+1FKiLXxHSkmPseP+kNlulaMuP3n2brvKWEqk/Jc4=
 github.com/hashicorp/go-multierror v1.1.1 h1:H5DkEtf6CXdFp0N0Em5UCwQpXMWke8IA0+lD48awMYo=
@@ -308,6 +308,8 @@ github.com/sergi/go-diff v1.2.0 h1:XU+rvMAioB0UC3q1MFrIQy4Vo5/4VsRDQQXHsEya6xQ=
 github.com/sergi/go-diff v1.2.0/go.mod h1:STckp+ISIX8hZLjrqAeVduY0gWCT9IjLuqbuNXdaHfM=
 github.com/sirupsen/logrus v1.9.3 h1:dueUQJ1C2q9oE3F7wvmSGAaVtTmUizReu6fjN8uqzbQ=
 github.com/sirupsen/logrus v1.9.3/go.mod h1:naHLuLoDiP4jHNo9R0sCBMtWGeIprob74mVsIT4qYEQ=
+github.com/sourcegraph/go-diff-patch v0.0.0-20240223163233-798fd1e94a8e h1:H+jDTUeF+SVd4ApwnSFoew8ZwGNRfgb9EsZc7LcocAg=
+github.com/sourcegraph/go-diff-patch v0.0.0-20240223163233-798fd1e94a8e/go.mod h1:VsUklG6OQo7Ctunu0gS3AtEOCEc2kMB6r5rKzxAes58=
 github.com/spf13/cobra v1.8.0 h1:7aJaZx1B85qltLMc546zn58BxxfZdR/W22ej9CFoEf0=
 github.com/spf13/cobra v1.8.0/go.mod h1:WXLWApfZ71AjXPya3WOlMsY9yMs7YeiHhFVlvLyhcho=
 github.com/spf13/pflag v1.0.5 h1:iy+VFUOCP1a+8yFto/drg2CJ5u0yRoB7fZw3DKv/JXA=

pkg/builtin/builtin.go

@@ -29,6 +29,7 @@ var SafeTools = map[string]struct{}{
 	"sys.abort":        {},
 	"sys.chat.finish":  {},
 	"sys.chat.history": {},
+	"sys.chat.current": {},
 	"sys.echo":         {},
 	"sys.prompt":       {},
 	"sys.time.now":     {},
@@ -229,6 +230,15 @@ var tools = map[string]types.Tool{
 			BuiltinFunc: SysChatHistory,
 		},
 	},
+	"sys.chat.current": {
+		ToolDef: types.ToolDef{
+			Parameters: types.Parameters{
+				Description: "Retrieves the current chat dialog",
+				Arguments:   types.ObjectSchema(),
+			},
+			BuiltinFunc: SysChatCurrent,
+		},
+	},
 	"sys.context": {
 		ToolDef: types.ToolDef{
 			Parameters: types.Parameters{
@@ -715,6 +725,28 @@ func writeHistory(ctx *engine.Context) (result []engine.ChatHistoryCall) {
 	return
 }
 
+func SysChatCurrent(ctx context.Context, _ []string, _ string, _ chan<- string) (string, error) {
+	engineContext, _ := engine.FromContext(ctx)
+
+	var call any
+	if engineContext != nil && engineContext.CurrentReturn != nil && engineContext.CurrentReturn.State != nil {
+		call = engine.ChatHistoryCall{
+			ID:         engineContext.ID,
+			Tool:       engineContext.Tool,
+			Completion: engineContext.CurrentReturn.State.Completion,
+		}
+	} else {
+		call = map[string]any{}
+	}
+
+	data, err := json.Marshal(call)
+	if err != nil {
+		return invalidArgument("", err), nil
+	}
+
+	return string(data), nil
+}
+
 func SysChatFinish(_ context.Context, _ []string, input string, _ chan<- string) (string, error) {
 	var params struct {
 		Message string `json:"return,omitempty"`

pkg/cli/gptscript.go

@@ -467,7 +467,7 @@ func (r *GPTScript) Run(cmd *cobra.Command, args []string) (retErr error) {
 				DefaultModel:        r.DefaultModel,
 				TrustedRepoPrefixes: []string{"github.com/gptscript-ai"},
 				DisableCache:        r.DisableCache,
-				Input:               strings.Join(args[1:], " "),
+				Input:               toolInput,
 				CacheDir:            r.CacheDir,
 				SubTool:             r.SubTool,
 				Workspace:           r.Workspace,

pkg/engine/engine.go

@@ -59,6 +59,7 @@ type CallResult struct {
 type commonContext struct {
 	ID           string                `json:"id"`
 	Tool         types.Tool            `json:"tool"`
+	CurrentAgent types.ToolReference   `json:"currentAgent,omitempty"`
 	AgentGroup   []types.ToolReference `json:"agentGroup,omitempty"`
 	InputContext []InputContext        `json:"inputContext"`
 	ToolCategory ToolCategory          `json:"toolCategory,omitempty"`
@@ -73,10 +74,11 @@ type CallContext struct {
 
 type Context struct {
 	commonContext
-	Ctx        context.Context
-	Parent     *Context
-	LastReturn *Return
-	Program    *types.Program
+	Ctx           context.Context
+	Parent        *Context
+	LastReturn    *Return
+	CurrentReturn *Return
+	Program       *types.Program
 	// Input is saved only so that we can render display text, don't use otherwise
 	Input string
 }
@@ -129,6 +131,18 @@ func (c *Context) ParentID() string {
 	return c.Parent.ID
 }
 
+func (c *Context) CurrentAgent() types.ToolReference {
+	for _, ref := range c.AgentGroup {
+		if ref.ToolID == c.Tool.ID {
+			return ref
+		}
+	}
+	if c.Parent != nil {
+		return c.Parent.CurrentAgent()
+	}
+	return types.ToolReference{}
+}
+
 func (c *Context) GetCallContext() *CallContext {
 	var toolName string
 	if c.Parent != nil {
@@ -143,12 +157,15 @@ func (c *Context) GetCallContext() *CallContext {
 		}
 	}
 
-	return &CallContext{
+	result := &CallContext{
 		commonContext: c.commonContext,
 		ParentID:      c.ParentID(),
 		ToolName:      toolName,
 		DisplayText:   types.ToDisplayText(c.Tool, c.Input),
 	}
+
+	result.CurrentAgent = c.CurrentAgent()
+	return result
 }
 
 func (c *Context) UnmarshalJSON([]byte) error {
@@ -215,10 +232,11 @@ func (c *Context) SubCallContext(ctx context.Context, input, toolID, callID stri
 			AgentGroup:   agentGroup,
 			ToolCategory: toolCategory,
 		},
-		Ctx:     ctx,
-		Parent:  c,
-		Program: c.Program,
-		Input:   input,
+		Ctx:           ctx,
+		Parent:        c,
+		Program:       c.Program,
+		CurrentReturn: c.CurrentReturn,
+		Input:         input,
 	}, nil
 }
 
@@ -270,6 +288,7 @@ func (e *Engine) Start(ctx Context, input string) (ret *Return, _ error) {
 		MaxTokens:            tool.Parameters.MaxTokens,
 		JSONResponse:         tool.Parameters.JSONResponse,
 		Cache:                tool.Parameters.Cache,
+		Chat:                 tool.Parameters.Chat,
 		Temperature:          tool.Parameters.Temperature,
 		InternalSystemPrompt: tool.Parameters.InternalPrompt,
 	}

pkg/openai/client.go

@@ -300,11 +300,16 @@ func (c *Client) Call(ctx context.Context, messageRequest types.CompletionReques
 	if messageRequest.Model == "" {
 		messageRequest.Model = c.defaultModel
 	}
+
 	msgs, err := toMessages(messageRequest, !c.setSeed)
 	if err != nil {
 		return nil, err
 	}
 
+	if messageRequest.Chat {
+		msgs = dropMessagesOverCount(messageRequest.MaxTokens, msgs)
+	}
+
 	if len(msgs) == 0 {
 		log.Errorf("invalid request, no messages to send to LLM")
 		return &types.CompletionMessage{

pkg/openai/count.go

@@ -0,0 +1,57 @@
+package openai
+
+import openai "github.com/gptscript-ai/chat-completion-client"
+
+func dropMessagesOverCount(maxTokens int, msgs []openai.ChatCompletionMessage) (result []openai.ChatCompletionMessage) {
+	var (
+		lastSystem   int
+		withinBudget int
+		budget       = maxTokens
+	)
+
+	if maxTokens == 0 {
+		budget = 300_000
+	} else {
+		budget *= 3
+	}
+
+	for i, msg := range msgs {
+		if msg.Role == openai.ChatMessageRoleSystem {
+			budget -= countMessage(msg)
+			lastSystem = i
+			result = append(result, msg)
+		} else {
+			break
+		}
+	}
+
+	for i := len(msgs) - 1; i > lastSystem; i-- {
+		withinBudget = i
+		budget -= countMessage(msgs[i])
+		if budget <= 0 {
+			break
+		}
+	}
+
+	if withinBudget == len(msgs)-1 {
+		// We are going to drop all non system messages, which seems useless, so just return them
+		// all and let it fail
+		return msgs
+	}
+
+	return append(result, msgs[withinBudget:]...)
+}
+
+func countMessage(msg openai.ChatCompletionMessage) (count int) {
+	count += len(msg.Role)
+	count += len(msg.Content)
+	for _, content := range msg.MultiContent {
+		count += len(content.Text)
+	}
+	for _, tool := range msg.ToolCalls {
+		count += len(tool.Function.Name)
+		count += len(tool.Function.Arguments)
+	}
+	count += len(msg.ToolCallID)
+	return count / 3
+}

pkg/runner/output.go

@@ -41,7 +41,6 @@ func (r *Runner) handleOutput(callCtx engine.Context, monitor Monitor, env []str
 	for _, outputToolRef := range outputToolRefs {
 		inputData, err := json.Marshal(map[string]any{
 			"output":       output,
-			"chatFinish":   chatFinish,
 			"continuation": continuation,
 			"chat":         callCtx.Tool.Chat,
 		})

pkg/runner/runner.go

@@ -559,6 +559,8 @@ func (r *Runner) resume(callCtx engine.Context, monitor Monitor, env []string, s
 	}
 
 	for {
+		callCtx.CurrentReturn = state.Continuation
+
 		if state.Continuation.Result != nil && len(state.Continuation.Calls) == 0 && state.SubCallID == "" && state.ResumeInput == nil {
 			progressClose()
 			monitor.Event(Event{

pkg/tests/runner_test.go

@@ -413,7 +413,8 @@ func TestSubChat(t *testing.T) {
                     ],
                     "usage": {}
                   }
-                ]
+                ],
+                "chat": true
               }
             },
             "result": "Assistant 1"
@@ -555,7 +556,8 @@ func TestSubChat(t *testing.T) {
                     ],
                     "usage": {}
                   }
-                ]
+                ],
+                "chat": true
               }
             },
             "result": "Assistant 2"
@@ -622,7 +624,8 @@ func TestChat(t *testing.T) {
               ],
               "usage": {}
             }
-          ]
+          ],
+          "chat": true
         }
       },
       "result": "Assistant 1"
@@ -691,7 +694,8 @@ func TestChat(t *testing.T) {
               ],
               "usage": {}
             }
-          ]
+          ],
+          "chat": true
         }
       },
       "result": "Assistant 2"
@@ -866,7 +870,7 @@ func TestOutput(t *testing.T) {
 	require.NoError(t, err)
 	r.AssertResponded(t)
 	assert.False(t, resp.Done)
-	autogold.Expect(`CHAT: true CONTENT: Response 1 CONTINUATION: true FINISH: false suffix
+	autogold.Expect(`CHAT: true CONTENT: Response 1 CONTINUATION: true suffix
 `).Equal(t, resp.Content)
 	autogold.ExpectFile(t, toJSONString(t, resp), autogold.Name(t.Name()+"/step1"))
 
@@ -877,7 +881,7 @@ func TestOutput(t *testing.T) {
 	require.NoError(t, err)
 	r.AssertResponded(t)
 	assert.False(t, resp.Done)
-	autogold.Expect(`CHAT: true CONTENT: Response 2 CONTINUATION: true FINISH: false suffix
+	autogold.Expect(`CHAT: true CONTENT: Response 2 CONTINUATION: true suffix
 `).Equal(t, resp.Content)
 	autogold.ExpectFile(t, toJSONString(t, resp), autogold.Name(t.Name()+"/step2"))
 
@@ -890,7 +894,7 @@ func TestOutput(t *testing.T) {
 	require.NoError(t, err)
 	r.AssertResponded(t)
 	assert.True(t, resp.Done)
-	autogold.Expect(`CHAT FINISH: CHAT: true CONTENT: Chat Done CONTINUATION: false FINISH: true suffix
+	autogold.Expect(`CHAT FINISH: CHAT: true CONTENT: Chat Done CONTINUATION: false suffix
 `).Equal(t, resp.Content)
 	autogold.ExpectFile(t, toJSONString(t, resp), autogold.Name(t.Name()+"/step3"))
 }

pkg/tests/testdata/TestAgents/call1.golden

@@ -52,5 +52,6 @@
       ],
       "usage": {}
     }
-  ]
+  ],
+  "chat": true
 }`

pkg/tests/testdata/TestAgents/call2.golden

@@ -28,5 +28,6 @@
       ],
       "usage": {}
     }
-  ]
+  ],
+  "chat": true
 }`

pkg/tests/testdata/TestAgents/call3.golden

@@ -43,5 +43,6 @@
       ],
       "usage": {}
     }
-  ]
+  ],
+  "chat": true
 }`

pkg/tests/testdata/TestAgents/call4.golden

@@ -11,5 +11,6 @@
       ],
       "usage": {}
     }
-  ]
+  ],
+  "chat": true
 }`