feat: add history messages, reasoning content, and adaptive language support

- Add optional history parameter to Run/RunStream for conversation context
- Add Reasoning/ReasoningDelta fields to completion and stream types
- Update llama.cpp adapter to propagate reasoning content from responses
- Default persona language now adapts to the user's language dynamically

docs/architecture.es.md

@@ -206,8 +206,8 @@ Bucle iterativo entre LLM y ejecución de tools. Es el "cerebro" que orquesta to
 
 ```go
 type Loop interface {
-    Run(ctx context.Context, input string) (Response, error)
-    RunStream(ctx context.Context, input string) iter.Seq2[Chunk, error]
+    Run(ctx context.Context, input string, history ...Message) (Response, error)
+    RunStream(ctx context.Context, input string, history ...Message) iter.Seq2[Chunk, error]
 }
 
 type Config struct {

docs/architecture.md

@@ -206,8 +206,8 @@ Iterative loop between LLM and tool execution. It's the "brain" that orchestrate
 
 ```go
 type Loop interface {
-    Run(ctx context.Context, input string) (Response, error)
-    RunStream(ctx context.Context, input string) iter.Seq2[Chunk, error]
+    Run(ctx context.Context, input string, history ...Message) (Response, error)
+    RunStream(ctx context.Context, input string, history ...Message) iter.Seq2[Chunk, error]
 }
 
 type Config struct {

pkg/agent/README.es.md

@@ -23,8 +23,8 @@ while iteration < MaxIterations:
 
 ```go
 type Loop interface {
-    Run(ctx context.Context, input string) (Response, error)
-    RunStream(ctx context.Context, input string) iter.Seq2[Chunk, error]
+    Run(ctx context.Context, input string, history ...llm.Message) (Response, error)
+    RunStream(ctx context.Context, input string, history ...llm.Message) iter.Seq2[Chunk, error]
 }
 
 type Config struct {

pkg/agent/README.md

@@ -20,8 +20,8 @@ while iteration < MaxIterations:
 
 ```go
 type Loop interface {
-    Run(ctx context.Context, input string) (Response, error)
-    RunStream(ctx context.Context, input string) iter.Seq2[Chunk, error]
+    Run(ctx context.Context, input string, history ...llm.Message) (Response, error)
+    RunStream(ctx context.Context, input string, history ...llm.Message) iter.Seq2[Chunk, error]
 }
 
 type Config struct {

pkg/agent/loop.go

@@ -75,10 +75,12 @@ func New(cfg Config) *Loop {
 }
 
 // Run executes the agent loop and returns the final response.
-func (l *Loop) Run(ctx context.Context, input string) (Response, error) {
+// Optional history messages are appended after the system prompt and before
+// the new user input.
+func (l *Loop) Run(ctx context.Context, input string, history ...llm.Message) (Response, error) {
 	start := time.Now()
 
-	messages := l.buildInitialMessages(input)
+	messages := l.buildInitialMessages(input, history)
 	var finalContent string
 	var allToolCalls []llm.ToolCall
 	var totalUsage llm.TokenUsage
@@ -137,9 +139,11 @@ func (l *Loop) Run(ctx context.Context, input string) (Response, error) {
 }
 
 // RunStream executes the agent loop with streaming output.
-func (l *Loop) RunStream(ctx context.Context, input string) iter.Seq2[llm.StreamChunk, error] {
+// Optional history messages are appended after the system prompt and before
+// the new user input.
+func (l *Loop) RunStream(ctx context.Context, input string, history ...llm.Message) iter.Seq2[llm.StreamChunk, error] {
 	return func(yield func(llm.StreamChunk, error) bool) {
-		messages := l.buildInitialMessages(input)
+		messages := l.buildInitialMessages(input, history)
 		iterations := 0
 
 		for iterations < l.cfg.MaxIters {
@@ -177,7 +181,7 @@ func (l *Loop) RunStream(ctx context.Context, input string) iter.Seq2[llm.Stream
 					}
 				}
 
-				if !hasToolCalls && chunk.Delta != "" {
+				if !hasToolCalls && (chunk.Delta != "" || chunk.ReasoningDelta != "") {
 					responseBuilder.WriteString(chunk.Delta)
 					if !yield(chunk, nil) {
 						return
@@ -194,12 +198,13 @@ func (l *Loop) RunStream(ctx context.Context, input string) iter.Seq2[llm.Stream
 	}
 }
 
-func (l *Loop) buildInitialMessages(input string) []llm.Message {
+func (l *Loop) buildInitialMessages(input string, history []llm.Message) []llm.Message {
 	systemPrompt := persona.AssembleSystemPrompt(l.cfg.Persona, "")
-	return []llm.Message{
-		{Role: llm.RoleSystem, Content: systemPrompt},
-		{Role: llm.RoleUser, Content: input},
-	}
+	messages := make([]llm.Message, 0, len(history)+2)
+	messages = append(messages, llm.Message{Role: llm.RoleSystem, Content: systemPrompt})
+	messages = append(messages, history...)
+	messages = append(messages, llm.Message{Role: llm.RoleUser, Content: input})
+	return messages
 }
 
 func (l *Loop) getToolSchemas() []json.RawMessage {

pkg/llm/providers/llamacpp/client.go

@@ -125,6 +125,7 @@ func (c *Client) Stream(ctx context.Context, req llm.CompletionRequest) iter.Seq
 			for _, choice := range event.Choices {
 				chunk := llm.StreamChunk{
 					Delta:          choice.Delta.Content,
+					ReasoningDelta: choice.Delta.ReasoningContent,
 				}
 				if choice.FinishReason != "" {
 					chunk.FinishReason = choice.FinishReason
@@ -210,6 +211,7 @@ func (c *Client) toResponse(resp llamaChatResponse) llm.CompletionResponse {
 		ID:         resp.ID,
 		Model:      resp.Model,
 		Content:    choice.Message.Content,
+		Reasoning:  choice.Message.ReasoningContent,
 		StopReason: choice.FinishReason,
 	}
 
@@ -272,6 +274,7 @@ type llamaChoice struct {
 type llamaMessageResult struct {
 	Role             string          `json:"role"`
 	Content          string          `json:"content"`
+	ReasoningContent string          `json:"reasoning_content"`
 	ToolCalls        []llamaToolCall `json:"tool_calls"`
 }
 
@@ -307,6 +310,7 @@ type llamaStreamChoice struct {
 
 type llamaStreamDelta struct {
 	Content          string                `json:"content"`
+	ReasoningContent string                `json:"reasoning_content"`
 	Role              string               `json:"role"`
 	ToolCalls         []llamaStreamToolCall `json:"tool_calls"`
 }

pkg/llm/types.go

@@ -82,6 +82,7 @@ type CompletionResponse struct {
 	ID         string     `json:"id"`
 	Model      string     `json:"model"`
 	Content    string     `json:"content"`
+	Reasoning  string     `json:"reasoning,omitempty"`
 	ToolCalls  []ToolCall `json:"tool_calls,omitempty"`
 	StopReason string     `json:"stop_reason"`
 	Usage      TokenUsage `json:"usage,omitempty"`
@@ -106,6 +107,7 @@ type ToolCall struct {
 // StreamChunk is emitted by the iterator returned from Stream().
 type StreamChunk struct {
 	Delta          string     `json:"delta"`
+	ReasoningDelta string     `json:"reasoning_delta,omitempty"`
 	ToolCalls      []ToolCall `json:"tool_calls,omitempty"`
 	FinishReason   string     `json:"finish_reason,omitempty"`
 	Usage          TokenUsage `json:"usage,omitempty"` // only present on final chunk

pkg/persona/persona.go

@@ -37,7 +37,7 @@ func DefaultPersona() Persona {
 		Name:     "Rony",
 		Tone:     "professional and helpful",
 		Style:    "clear and concise",
-		Language: "en",
+		Language: "the user's language",
 	}
 }
 
@@ -59,8 +59,12 @@ func AssembleSystemPrompt(p Persona, agentsMD string) string {
 		parts = append(parts, fmt.Sprintf("Write in a %s style.", p.Style))
 	}
 	if p.Language != "" {
+		if strings.EqualFold(p.Language, "the user's language") {
+			parts = append(parts, "Respond in the same language as the user's messages.")
+		} else {
 			parts = append(parts, fmt.Sprintf("Respond in %s.", p.Language))
 		}
+	}
 	for _, c := range p.Constraints {
 		parts = append(parts, fmt.Sprintf("CONSTRAINT: %s", c))
 	}