feat: implement tool registry and add various tools for HTTP, file operations, SSH, and Matrix messaging

agents/runtime.go

@@ -18,8 +18,11 @@ import (
 	shelllm "github.com/enmanuel/agents/shell/llm"
 	"github.com/enmanuel/agents/shell/matrix"
 	"github.com/enmanuel/agents/shell/ssh"
+	"github.com/enmanuel/agents/tools"
 )
 
+const defaultMaxToolIterations = 5
+
 // Agent is the assembled runtime: pure core + impure shell.
 type Agent struct {
 	cfg         *config.AgentConfig
@@ -29,6 +32,7 @@ type Agent struct {
 	matrix      *matrix.Client
 	runner      *effects.Runner
 	listener    *matrix.Listener
+	toolReg     *tools.Registry
 	logger      *slog.Logger
 	cryptoStore io.Closer // non-nil when E2EE is enabled; closed on shutdown
 }
@@ -75,12 +79,16 @@ func New(cfg *config.AgentConfig, rules []decision.Rule, logger *slog.Logger) (*
 	// Effects runner
 	runner := effects.NewRunner(matrixClient, sshExec, logger)
 
+	// Tool registry — register tools enabled in config
+	toolReg := buildToolRegistry(cfg, sshExec, matrixClient, logger)
+
 	a := &Agent{
 		cfg:         cfg,
 		rules:       rules,
 		llm:         llmFunc,
 		matrix:      matrixClient,
 		runner:      runner,
+		toolReg:     toolReg,
 		logger:      logger,
 		cryptoStore: cryptoStore,
 	}
@@ -96,7 +104,11 @@ func (a *Agent) Run(ctx context.Context) error {
 	if a.cryptoStore != nil {
 		defer a.cryptoStore.Close()
 	}
-	a.logger.Info("agent starting", "id", a.cfg.Agent.ID, "name", a.cfg.Agent.Name)
+	a.logger.Info("agent starting",
+		"id", a.cfg.Agent.ID,
+		"name", a.cfg.Agent.Name,
+		"tools", a.toolReg.Names(),
+	)
 	return a.listener.Run(ctx)
 }
 
@@ -134,7 +146,7 @@ func (a *Agent) handleEvent(ctx context.Context, msgCtx decision.MessageContext,
 		return
 	}
 
-	// Expand LLM actions inline (simplified — real impl would maintain conversation state)
+	// Expand LLM actions inline — with tool-use loop when enabled
 	expanded := make([]decision.Action, 0, len(actions))
 	for _, act := range actions {
 		if act.Kind == decision.ActionKindLLM {
@@ -164,20 +176,120 @@ func (a *Agent) runLLM(ctx context.Context, msgCtx decision.MessageContext) (str
 		"model", a.cfg.LLM.Primary.Model,
 		"provider", a.cfg.LLM.Primary.Provider,
 	)
-	req := coretypes.CompletionRequest{
-		Model:        a.cfg.LLM.Primary.Model,
-		MaxTokens:    a.cfg.LLM.Primary.MaxTokens,
-		Temperature:  a.cfg.LLM.Primary.Temperature,
-		SystemPrompt: a.cfg.Agent.Description,
-		Messages: []coretypes.Message{
-			{Role: coretypes.RoleUser, Content: msgCtx.Content},
-		},
+
+	// Load system prompt from file if configured, else use description
+	systemPrompt := a.cfg.Agent.Description
+
+	messages := []coretypes.Message{
+		{Role: coretypes.RoleUser, Content: msgCtx.Content},
 	}
-	resp, err := a.llm(ctx, req)
-	if err != nil {
-		a.logger.Error("LLM call failed", "model", req.Model, "err", err)
-		return "", err
+
+	// Build tool specs for the LLM if tool_use is enabled
+	var llmTools []coretypes.ToolSpec
+	if a.cfg.LLM.ToolUse.Enabled && a.toolReg.Len() > 0 {
+		llmTools = a.toolReg.ToLLMSpecs()
+		a.logger.Debug("tools available for LLM", "count", len(llmTools))
 	}
-	a.logger.Debug("LLM responded", "content_len", len(resp.Content))
-	return resp.Content, nil
+
+	maxIter := a.cfg.LLM.ToolUse.MaxIterations
+	if maxIter <= 0 {
+		maxIter = defaultMaxToolIterations
+	}
+
+	// Tool-use loop: call LLM → execute tools → feed results back → repeat
+	for i := 0; i < maxIter; i++ {
+		req := coretypes.CompletionRequest{
+			Model:        a.cfg.LLM.Primary.Model,
+			MaxTokens:    a.cfg.LLM.Primary.MaxTokens,
+			Temperature:  a.cfg.LLM.Primary.Temperature,
+			SystemPrompt: systemPrompt,
+			Messages:     messages,
+			Tools:        llmTools,
+		}
+
+		resp, err := a.llm(ctx, req)
+		if err != nil {
+			a.logger.Error("LLM call failed", "model", req.Model, "err", err)
+			return "", err
+		}
+
+		a.logger.Debug("LLM responded",
+			"content_len", len(resp.Content),
+			"tool_calls", len(resp.ToolCalls),
+			"finish_reason", resp.FinishReason,
+		)
+
+		// No tool calls — return the text response
+		if len(resp.ToolCalls) == 0 {
+			return resp.Content, nil
+		}
+
+		// Append assistant message with tool calls to conversation
+		messages = append(messages, coretypes.Message{
+			Role:      coretypes.RoleAssistant,
+			Content:   resp.Content,
+			ToolCalls: resp.ToolCalls,
+		})
+
+		// Execute each tool and append results
+		for _, tc := range resp.ToolCalls {
+			a.logger.Info("executing tool",
+				"tool", tc.Name,
+				"call_id", tc.ID,
+			)
+
+			result := a.toolReg.Execute(ctx, tc.Name, tc.Arguments)
+
+			output := result.Output
+			if result.Err != nil {
+				output = fmt.Sprintf("error: %s", result.Err)
+				a.logger.Warn("tool execution error",
+					"tool", tc.Name,
+					"err", result.Err,
+				)
+			} else {
+				a.logger.Debug("tool executed",
+					"tool", tc.Name,
+					"output_len", len(output),
+				)
+			}
+
+			messages = append(messages, coretypes.Message{
+				Role:       coretypes.RoleTool,
+				Content:    output,
+				ToolCallID: tc.ID,
+			})
+		}
+	}
+
+	// Max iterations reached — return whatever we have
+	a.logger.Warn("tool-use loop reached max iterations", "max", maxIter)
+	return "I've reached the maximum number of tool iterations. Here's what I found so far.", nil
+}
+
+// buildToolRegistry creates a Registry with tools enabled in the agent's config.
+func buildToolRegistry(cfg *config.AgentConfig, sshExec *ssh.Executor, matrixClient *matrix.Client, logger *slog.Logger) *tools.Registry {
+	reg := tools.NewRegistry()
+
+	if cfg.Tools.HTTP.Enabled {
+		reg.Register(tools.NewHTTPGet(cfg.Tools.HTTP))
+		reg.Register(tools.NewHTTPPost(cfg.Tools.HTTP))
+		logger.Debug("registered http tools")
+	}
+
+	if cfg.Tools.SSH.Enabled {
+		reg.Register(tools.NewSSHCommand(cfg.Tools.SSH, sshExec))
+		logger.Debug("registered ssh tool")
+	}
+
+	if cfg.Tools.FileOps.Enabled {
+		reg.Register(tools.NewReadFile(cfg.Tools.FileOps))
+		logger.Debug("registered file tool")
+	}
+
+	// matrix_send is always available
+	reg.Register(tools.NewMatrixSend(matrixClient))
+	logger.Debug("registered matrix tool")
+
+	return reg
 }