From 1c3ea7d1f1b225491eff97428afeb660ee5e684d Mon Sep 17 00:00:00 2001 From: Steve Dudenhoeffer Date: Sat, 3 May 2025 22:09:02 -0400 Subject: [PATCH] Refactor Knowledge struct into shared package Moved the Knowledge struct and related types to the shared package, updating all references across the codebase. This improves modularity and enables better reuse of the Knowledge type across different components. --- cmd/console/cmd.go | 3 ++- pkg/agents/console-new/agent.go | 8 ++++---- pkg/agents/console.go | 4 ++-- pkg/agents/extract_knowledge.go | 8 +++++--- pkg/agents/knowledge_integrate.go | 8 +++++--- pkg/agents/knowledge_processor.go | 4 +++- pkg/agents/read_page.go | 10 ++++++---- pkg/agents/search.go | 14 ++++++++------ pkg/agents/{ => shared}/knowledge.go | 2 +- pkg/agents/shared/knowledgeworker.go | 26 ++++++++++++-------------- pkg/agents/youtube.go | 27 +++++++++++++++------------ 11 files changed, 63 insertions(+), 51 deletions(-) rename pkg/agents/{ => shared}/knowledge.go (99%) diff --git a/cmd/console/cmd.go b/cmd/console/cmd.go index 2fc6d01..34a83e3 100644 --- a/cmd/console/cmd.go +++ b/cmd/console/cmd.go @@ -9,6 +9,7 @@ import ( "time" "gitea.stevedudenhoeffer.com/steve/answer/pkg/agents" + "gitea.stevedudenhoeffer.com/steve/answer/pkg/agents/shared" "gitea.stevedudenhoeffer.com/steve/answer/pkg/agents/console" @@ -102,7 +103,7 @@ func main() { c := console.Agent{ Model: m, - OnDone: func(ctx context.Context, knowledge agents.Knowledge) error { return nil }, + OnDone: func(ctx context.Context, knowledge shared.Knowledge) error { return nil }, OnCommandStart: func(ctx context.Context, command string) error { slog.Info("command", "command", command) return nil diff --git a/pkg/agents/console-new/agent.go b/pkg/agents/console-new/agent.go index 3b0960f..9c3a391 100644 --- a/pkg/agents/console-new/agent.go +++ b/pkg/agents/console-new/agent.go @@ -23,12 +23,12 @@ type Agent struct { // Model is the chat completion model to use Model gollm.ChatCompletion - OnLoopComplete func(ctx context.Context, knowledge agents.Knowledge) error + OnLoopComplete func(ctx context.Context, knowledge shared.Knowledge) error OnCommandStart func(ctx context.Context, command string) error OnCommandDone func(ctx context.Context, command string, output string, err error) error - OnDone func(ctx context.Context, knowledge agents.Knowledge) error + OnDone func(ctx context.Context, knowledge shared.Knowledge) error ContextualInformation []string @@ -36,7 +36,7 @@ type Agent struct { } type Response struct { - Knowledge agents.Knowledge + Knowledge shared.Knowledge DataDir string OutputDir string } @@ -52,7 +52,7 @@ func (a Agent) Answer(ctx context.Context, questions []string) (Response, error) a.MaxCommands = 20 // Default to 20 commands as per requirements } - res.Knowledge = agents.Knowledge{ + res.Knowledge = shared.Knowledge{ OriginalQuestions: questions, RemainingQuestions: questions, } diff --git a/pkg/agents/console.go b/pkg/agents/console.go index a519b7c..39a69d0 100644 --- a/pkg/agents/console.go +++ b/pkg/agents/console.go @@ -24,13 +24,13 @@ type ConsoleConfig struct { OnCommandDone func(ctx context.Context, command string, output string, err error) error } -func (a Agent) Console(ctx context.Context, questions []string, cfg ConsoleConfig) (Knowledge, string, error) { +func (a Agent) Console(ctx context.Context, questions []string, cfg ConsoleConfig) (shared.Knowledge, string, error) { if cfg.MaxCommands <= 0 { cfg.MaxCommands = 10000 } - var resKnowledge = Knowledge{ + var resKnowledge = shared.Knowledge{ OriginalQuestions: questions, RemainingQuestions: questions, } diff --git a/pkg/agents/extract_knowledge.go b/pkg/agents/extract_knowledge.go index 2fa1200..1c86fbe 100644 --- a/pkg/agents/extract_knowledge.go +++ b/pkg/agents/extract_knowledge.go @@ -5,6 +5,8 @@ import ( "strings" gollm "gitea.stevedudenhoeffer.com/steve/go-llm" + + "gitea.stevedudenhoeffer.com/steve/answer/pkg/agents/shared" ) // ExtractKnowledge will take a knowledge object and use the gained knowledge to extract the knowledge relevant to the @@ -16,16 +18,16 @@ import ( // contextualInformation is any contextual information that should be provided to the model. // It will return the knowledge extracted from the sourceData along with any remaining questions. // This agent call will not use the Agent's system prompts, but will instead form its own. The contextual information will be used. -func (a Agent) ExtractKnowledge(ctx context.Context, sourceData string, source string, questions []string) (Knowledge, error) { +func (a Agent) ExtractKnowledge(ctx context.Context, sourceData string, source string, questions []string) (shared.Knowledge, error) { - var knowledge Knowledge + var knowledge shared.Knowledge fnAnswer := gollm.NewFunction( "learn", `Use learn to pass some relevant information to the model. The model will use this information to answer the question. Use it to learn relevant information from the text. Keep these concise and relevant to the question.`, func(ctx *gollm.Context, args struct { Info string `description:"The information to learn from the text."` }) (any, error) { - knowledge.Knowledge = append(knowledge.Knowledge, TidBit{Info: args.Info, Source: source}) + knowledge.Knowledge = append(knowledge.Knowledge, shared.TidBit{Info: args.Info, Source: source}) return "", nil }) diff --git a/pkg/agents/knowledge_integrate.go b/pkg/agents/knowledge_integrate.go index 0108032..2eab69b 100644 --- a/pkg/agents/knowledge_integrate.go +++ b/pkg/agents/knowledge_integrate.go @@ -4,12 +4,14 @@ import ( "context" gollm "gitea.stevedudenhoeffer.com/steve/go-llm" + + "gitea.stevedudenhoeffer.com/steve/answer/pkg/agents/shared" ) // KnowledgeIntegrate will ask the LLM to combine the gained knowledge with the current knowledge, and return the new representation of overall. // If source is not empty, then any new Knowledge will an empty source will be given the source. // This will override objectives, notes, and remaining questions. -func (a Agent) KnowledgeIntegrate(ctx context.Context, base Knowledge, in ...Knowledge) (Knowledge, error) { +func (a Agent) KnowledgeIntegrate(ctx context.Context, base shared.Knowledge, in ...shared.Knowledge) (shared.Knowledge, error) { // if there are no changes we can just return the knowledge if len(in) == 0 { return base, nil @@ -29,7 +31,7 @@ func (a Agent) KnowledgeIntegrate(ctx context.Context, base Knowledge, in ...Kno } } - var incoming Knowledge + var incoming shared.Knowledge for _, k := range in { incoming.NotesToSelf = append(incoming.NotesToSelf, k.NotesToSelf...) @@ -43,7 +45,7 @@ func (a Agent) KnowledgeIntegrate(ctx context.Context, base Knowledge, in ...Kno baseMsg.Text = "The original knowledge is as follows: " + baseMsg.Text incomingMsg.Text = "The new knowledge is as follows: " + incomingMsg.Text - var result = Knowledge{ + var result = shared.Knowledge{ OriginalQuestions: base.OriginalQuestions, Knowledge: append(base.Knowledge, incoming.Knowledge...), } diff --git a/pkg/agents/knowledge_processor.go b/pkg/agents/knowledge_processor.go index a101ea5..94a8602 100644 --- a/pkg/agents/knowledge_processor.go +++ b/pkg/agents/knowledge_processor.go @@ -8,10 +8,12 @@ import ( "strings" gollm "gitea.stevedudenhoeffer.com/steve/go-llm" + + "gitea.stevedudenhoeffer.com/steve/answer/pkg/agents/shared" ) // AnswerQuestionWithKnowledge will take a knowledge object and use the gained knowledge to answer a question. -func (a Agent) AnswerQuestionWithKnowledge(ctx context.Context, knowledge Knowledge) (string, error) { +func (a Agent) AnswerQuestionWithKnowledge(ctx context.Context, knowledge shared.Knowledge) (string, error) { originalQuestions := strings.Join(knowledge.OriginalQuestions, "\n") infoGained := "" diff --git a/pkg/agents/read_page.go b/pkg/agents/read_page.go index 29ccc19..62afd53 100644 --- a/pkg/agents/read_page.go +++ b/pkg/agents/read_page.go @@ -3,19 +3,21 @@ package agents import ( "context" "fmt" + "net/url" + + "gitea.stevedudenhoeffer.com/steve/answer/pkg/agents/shared" "gitea.stevedudenhoeffer.com/steve/answer/pkg/cache" "gitea.stevedudenhoeffer.com/steve/answer/pkg/extractor" - "net/url" ) -func (a Agent) ReadPage(ctx context.Context, u *url.URL, questions []string) (Knowledge, error) { +func (a Agent) ReadPage(ctx context.Context, u *url.URL, questions []string) (shared.Knowledge, error) { ar, err := extractArticle(ctx, u) if err != nil { - return Knowledge{}, err + return shared.Knowledge{}, err } if ar.Body == "" { - return Knowledge{}, fmt.Errorf("could not extract body from page") + return shared.Knowledge{}, fmt.Errorf("could not extract body from page") } return a.ExtractKnowledge(ctx, ar.Body, u.String(), questions) diff --git a/pkg/agents/search.go b/pkg/agents/search.go index 3172153..cb456af 100644 --- a/pkg/agents/search.go +++ b/pkg/agents/search.go @@ -14,6 +14,8 @@ import ( "gitea.stevedudenhoeffer.com/steve/go-extractor" "gitea.stevedudenhoeffer.com/steve/go-extractor/sites/duckduckgo" gollm "gitea.stevedudenhoeffer.com/steve/go-llm" + + "gitea.stevedudenhoeffer.com/steve/answer/pkg/agents/shared" ) func deferClose(c io.Closer) { @@ -26,7 +28,7 @@ func deferClose(c io.Closer) { type SearchTool struct { Name string Description string - Function func(ctx context.Context, src *url.URL, questions []string) (Knowledge, error) + Function func(ctx context.Context, src *url.URL, questions []string) (shared.Knowledge, error) } // SearchAndUseTools will search duckduckgo for the given question, and then ask the LLM to select a search result to @@ -41,8 +43,8 @@ type SearchTool struct { // will be combined and returned. // messages will be appended to all search results. The types of messages that can be appended are both string and // gollm.Message. -func (a Agent) SearchAndUseTools(ctx context.Context, searchQuery string, questions []string, loops int, allowConcurrent bool, maxReads int, tools []SearchTool, messages ...any) (Knowledge, error) { - var knowledge = Knowledge{ +func (a Agent) SearchAndUseTools(ctx context.Context, searchQuery string, questions []string, loops int, allowConcurrent bool, maxReads int, tools []SearchTool, messages ...any) (shared.Knowledge, error) { + var knowledge = shared.Knowledge{ OriginalQuestions: questions, RemainingQuestions: questions, } @@ -184,13 +186,13 @@ Use appropriate tools to analyze the search results and determine if they answer } slog.Info("search results called and executed", "error", err, "results text", results.Text, "results", results.CallResults) - var learned []Knowledge + var learned []shared.Knowledge for _, r := range results.CallResults { if r.Error != nil { continue } - if k, ok := r.Result.(Knowledge); ok { + if k, ok := r.Result.(shared.Knowledge); ok { learned = append(learned, k) } else { slog.Error("result is not knowledge", "result", r.Result) @@ -210,7 +212,7 @@ Use appropriate tools to analyze the search results and determine if they answer return knowledge, nil } -func (a Agent) SearchAndRead(ctx context.Context, searchQuery string, questions []string, allowConcurrent bool, maxReads int) (Knowledge, error) { +func (a Agent) SearchAndRead(ctx context.Context, searchQuery string, questions []string, allowConcurrent bool, maxReads int) (shared.Knowledge, error) { return a.SearchAndUseTools(ctx, searchQuery, questions, 2, allowConcurrent, maxReads, []SearchTool{ { Name: "readpage", diff --git a/pkg/agents/knowledge.go b/pkg/agents/shared/knowledge.go similarity index 99% rename from pkg/agents/knowledge.go rename to pkg/agents/shared/knowledge.go index 70961d7..ab5633b 100644 --- a/pkg/agents/knowledge.go +++ b/pkg/agents/shared/knowledge.go @@ -1,4 +1,4 @@ -package agents +package shared import ( "strings" diff --git a/pkg/agents/shared/knowledgeworker.go b/pkg/agents/shared/knowledgeworker.go index 77d99ab..cd37354 100644 --- a/pkg/agents/shared/knowledgeworker.go +++ b/pkg/agents/shared/knowledgeworker.go @@ -5,8 +5,6 @@ import ( "fmt" "strings" - "gitea.stevedudenhoeffer.com/steve/answer/pkg/agents" - gollm "gitea.stevedudenhoeffer.com/steve/go-llm" ) @@ -27,7 +25,7 @@ const DefaultPrompt = `Use the provided tools to answer the questions in your cu // source is the source of the knowledge, for example a URL. // Any tool call that returns a Knowledge object will be handled by this function in crafting the final Knowledge object. // Any other return type will be passed to the resultWorker function, if provided. -func (w KnowledgeWorker) Answer(context context.Context, knowledge *agents.Knowledge, systemPrompt string, userInput string, source string, history []gollm.Message, resultWorker func(res gollm.ToolCallResponse)) (agents.Knowledge, error) { +func (w KnowledgeWorker) Answer(context context.Context, knowledge *Knowledge, systemPrompt string, userInput string, source string, history []gollm.Message, resultWorker func(res gollm.ToolCallResponse)) (Knowledge, error) { var req gollm.Request if systemPrompt != "" { @@ -80,7 +78,7 @@ func (w KnowledgeWorker) Answer(context context.Context, knowledge *agents.Knowl func(ctx *gollm.Context, args struct { NotesToSelf []string `description:"Notes to leave for yourself for later."` }) (any, error) { - return agents.Knowledge{ + return Knowledge{ NotesToSelf: args.NotesToSelf, }, nil }), @@ -90,7 +88,7 @@ func (w KnowledgeWorker) Answer(context context.Context, knowledge *agents.Knowl func(ctx *gollm.Context, args struct { Objectives []string `description:"The objectives to set for executions going forward."` }) (any, error) { - return agents.Knowledge{ + return Knowledge{ CurrentObjectives: args.Objectives, }, nil }), @@ -100,13 +98,13 @@ func (w KnowledgeWorker) Answer(context context.Context, knowledge *agents.Knowl func(ctx *gollm.Context, args struct { Info []string `description:"The information to learn from the input."` }) (any, error) { - var k []agents.TidBit + var k []TidBit for _, i := range args.Info { - k = append(k, agents.TidBit{Info: i, Source: source}) + k = append(k, TidBit{Info: i, Source: source}) } - return agents.Knowledge{ + return Knowledge{ Knowledge: k, }, nil })). @@ -120,17 +118,17 @@ func (w KnowledgeWorker) Answer(context context.Context, knowledge *agents.Knowl resp, err := w.Model.ChatComplete(context, req) if err != nil { - return agents.Knowledge{}, fmt.Errorf("error calling model: %w", err) + return Knowledge{}, fmt.Errorf("error calling model: %w", err) } if len(resp.Choices) == 0 { - return agents.Knowledge{}, fmt.Errorf("no choices found") + return Knowledge{}, fmt.Errorf("no choices found") } choice := resp.Choices[0] if len(choice.Calls) == 0 { - return agents.Knowledge{}, fmt.Errorf("no calls found") + return Knowledge{}, fmt.Errorf("no calls found") } var callNames []string @@ -141,14 +139,14 @@ func (w KnowledgeWorker) Answer(context context.Context, knowledge *agents.Knowl results, err := w.ToolBox.ExecuteCallbacks(gollm.NewContext(context, req, &choice, nil), choice.Calls, w.OnNewFunction, w.OnFunctionFinished) if err != nil { - return agents.Knowledge{}, fmt.Errorf("error executing callbacks: %w", err) + return Knowledge{}, fmt.Errorf("error executing callbacks: %w", err) } - var res = agents.Knowledge{} + var res = Knowledge{} for _, r := range results { switch v := r.Result.(type) { - case agents.Knowledge: + case Knowledge: res = res.Absorb(v) default: diff --git a/pkg/agents/youtube.go b/pkg/agents/youtube.go index 64bcbd1..f8e03af 100644 --- a/pkg/agents/youtube.go +++ b/pkg/agents/youtube.go @@ -3,25 +3,28 @@ package agents import ( "context" "fmt" - "github.com/asticode/go-astisub" - "github.com/lrstanley/go-ytdlp" "io" "log/slog" "net/url" "os" "path/filepath" + + "github.com/asticode/go-astisub" + "github.com/lrstanley/go-ytdlp" + + "gitea.stevedudenhoeffer.com/steve/answer/pkg/agents/shared" ) func init() { ytdlp.MustInstall(context.Background(), nil) } -func (a Agent) ReadYouTubeTranscript(ctx context.Context, u *url.URL, questions []string) (Knowledge, error) { +func (a Agent) ReadYouTubeTranscript(ctx context.Context, u *url.URL, questions []string) (shared.Knowledge, error) { dlp := ytdlp.New() tmpDir, err := os.MkdirTemp("", "mort-ytdlp-") if err != nil { - return Knowledge{}, fmt.Errorf("error creating temp dir: %w", err) + return shared.Knowledge{}, fmt.Errorf("error creating temp dir: %w", err) } slog.Info("created temp dir", "path", tmpDir) @@ -40,15 +43,15 @@ func (a Agent) ReadYouTubeTranscript(ctx context.Context, u *url.URL, questions res, err := dlp.Run(ctx, u.String()) if err != nil { - return Knowledge{}, fmt.Errorf("error running yt-dlp: %w", err) + return shared.Knowledge{}, fmt.Errorf("error running yt-dlp: %w", err) } if res == nil { - return Knowledge{}, fmt.Errorf("yt-dlp returned nil") + return shared.Knowledge{}, fmt.Errorf("yt-dlp returned nil") } if res.ExitCode != 0 { - return Knowledge{}, fmt.Errorf("yt-dlp exited with code %d", res.ExitCode) + return shared.Knowledge{}, fmt.Errorf("yt-dlp exited with code %d", res.ExitCode) } // the transcript for this video now _should_ be at tmpDir/subs.en.vtt, however if it's not then just fine any @@ -60,7 +63,7 @@ func (a Agent) ReadYouTubeTranscript(ctx context.Context, u *url.URL, questions vttFile = "" files, err := os.ReadDir(tmpDir) if err != nil { - return Knowledge{}, fmt.Errorf("error reading directory: %w", err) + return shared.Knowledge{}, fmt.Errorf("error reading directory: %w", err) } for _, file := range files { @@ -72,7 +75,7 @@ func (a Agent) ReadYouTubeTranscript(ctx context.Context, u *url.URL, questions } if vttFile == "" { - return Knowledge{}, fmt.Errorf("no vtt file found") + return shared.Knowledge{}, fmt.Errorf("no vtt file found") } fp, err := os.Open(vttFile) @@ -83,16 +86,16 @@ func (a Agent) ReadYouTubeTranscript(ctx context.Context, u *url.URL, questions } }(fp) if err != nil { - return Knowledge{}, fmt.Errorf("error opening vtt file: %w", err) + return shared.Knowledge{}, fmt.Errorf("error opening vtt file: %w", err) } subs, err := astisub.ReadFromWebVTT(fp) if err != nil { - return Knowledge{}, fmt.Errorf("error reading vtt file: %w", err) + return shared.Knowledge{}, fmt.Errorf("error reading vtt file: %w", err) } if len(subs.Items) == 0 { - return Knowledge{}, fmt.Errorf("no subtitles found") + return shared.Knowledge{}, fmt.Errorf("no subtitles found") } var ts string