Add core implementation for AI-powered question answering

Introduce multiple agents, tools, and utilities for processing, extracting, and answering user-provided questions using LLMs and external data. Key features include knowledge processing, question splitting, search term generation, and contextual knowledge handling.

pkg/agents/remaining_questions.go

@@ -0,0 +1,101 @@
+package agents
+
+import (
+	"context"
+	"fmt"
+	"strings"
+
+	"gitea.stevedudenhoeffer.com/steve/answer/pkg/agents/shared"
+
+	gollm "gitea.stevedudenhoeffer.com/steve/go-llm"
+)
+
+type RemainingQuestions struct {
+	Model                 gollm.ChatCompletion
+	ContextualInformation []string
+}
+
+// Process takes a knowledge object and processes it into a response string.
+func (a RemainingQuestions) Process(ctx context.Context, knowledge shared.Knowledge) ([]string, error) {
+	originalQuestions := strings.Join(knowledge.OriginalQuestions, "\n")
+	infoGained := ""
+
+	// group all the gained knowledge by source
+	var m = map[string][]string{}
+	for _, k := range knowledge.Knowledge {
+		m[k.Source] = append(m[k.Source], k.Info)
+	}
+
+	// now order them in a list so they can be referenced by index
+	type source struct {
+		source string
+		info   []string
+	}
+
+	var sources []source
+	for k, v := range m {
+		sources = append(sources, source{
+			source: k,
+			info:   v,
+		})
+
+		if len(infoGained) > 0 {
+			infoGained += "\n"
+		}
+
+		infoGained += strings.Join(v, "\n")
+	}
+
+	systemPrompt := `I am trying to answer a question, and I gathered some knowledge in an attempt to do so. Here is what I am trying to answer: 
+` + originalQuestions + `
+
+Here is the knowledge I have gathered from ` + fmt.Sprint(len(sources)) + ` sources:
+` + infoGained
+
+	systemPrompt += "\n\nUsing the information gathered, have all of the questions been answered?  If not, what questions remain?  Use the function 'remaining_questions' to answer this question with 0 or more remaining questions."
+
+	var res []string
+	req := gollm.Request{
+		Messages: []gollm.Message{
+			{
+				Role: gollm.RoleSystem,
+				Text: systemPrompt,
+			},
+		},
+		Toolbox: gollm.NewToolBox(
+			gollm.NewFunction(
+				"remaining_questions",
+				"Given the information learned above, the following questions remain unanswered",
+				func(ctx *gollm.Context, args struct {
+					RemainingQuestions []string `description:"The questions that remain unanswered, if any"`
+				}) (string, error) {
+					res = append(res, args.RemainingQuestions...)
+					return "ok", nil
+				})),
+	}
+
+	if len(a.ContextualInformation) > 0 {
+		req.Messages = append(req.Messages, gollm.Message{
+			Role: gollm.RoleSystem,
+			Text: "Some contextual information you should be aware of: " + strings.Join(a.ContextualInformation, "\n"),
+		})
+	}
+
+	resp, err := a.Model.ChatComplete(ctx, req)
+	if err != nil {
+		return nil, fmt.Errorf("failed to chat complete: %w", err)
+	}
+
+	if len(resp.Choices) == 0 {
+		return nil, fmt.Errorf("no choices returned")
+	}
+
+	choice := resp.Choices[0]
+
+	if len(choice.Calls) == 0 {
+		return nil, fmt.Errorf("no calls returned")
+	}
+
+	_, err = req.Toolbox.ExecuteCallbacks(gollm.NewContext(ctx, req, &choice, nil), choice.Calls, nil, nil)
+	return res, err
+}