diff --git a/.golangci.yml b/.golangci.yml
index a9a6f07..9c9494b 100644
--- a/.golangci.yml
+++ b/.golangci.yml
@@ -48,7 +48,7 @@ linters:
       # More information: https://golangci-lint.run/usage/false-positives/#comments
       #
       # Please uncomment the following line if your code is not using the godoc format
-      - comments
+      # - comments
 
       # Common false positives
       # feel free to remove this if you don't have any false positives
@@ -126,6 +126,9 @@ linters:
         # Blank import should be only in a main or test package, or have a comment justifying it.
         - name: blank-imports
 
+        # Packages should have comments of the form "Package x ...".
+        - name: package-comments
+
         # context.Context() should be the first parameter of a function when provided as argument.
         - name: context-as-argument
           arguments:
diff --git a/Makefile b/Makefile
index 2ce49e2..c4ee60d 100644
--- a/Makefile
+++ b/Makefile
@@ -1,7 +1,7 @@
 BINARY_NAME=lambda
 TEST=simple
 
-.PHONY: help build run profile explain graph docs test bench clean
+.PHONY: help build run profile explain graph docs test bench lint clean
 .DEFAULT_GOAL := help
 .SILENT:
 
@@ -15,6 +15,7 @@ help:
 	echo "  docs     - Start local godoc server on port 6060"
 	echo "  test     - Run tests for all samples"
 	echo "  bench    - Run benchmarks for all samples"
+	echo "  lint     - Run golangci-lint on all packages"
 	echo "  clean    - Remove all build artifacts"
 
 build:
@@ -45,6 +46,9 @@ test:
 bench:
 	go test -bench=. -benchtime=10x -cpu=4 ./cmd/lambda
 
+lint:
+	go run github.com/golangci/golangci-lint/v2/cmd/golangci-lint@latest run ./...
+
 clean:
 	rm -f ${BINARY_NAME}
 	rm -f program.out
diff --git a/cmd/lambda/lambda.go b/cmd/lambda/lambda.go
index 7722471..60070fc 100644
--- a/cmd/lambda/lambda.go
+++ b/cmd/lambda/lambda.go
@@ -3,65 +3,29 @@ package main
 import (
 	"os"
 
-	"git.maximhutz.com/max/lambda/internal/cli"
-	"git.maximhutz.com/max/lambda/internal/config"
-	"git.maximhutz.com/max/lambda/internal/plugins"
-	"git.maximhutz.com/max/lambda/pkg/convert"
-	"git.maximhutz.com/max/lambda/pkg/normalorder"
-	"git.maximhutz.com/max/lambda/pkg/saccharine"
+	"github.com/spf13/cobra"
 )
 
-func main() {
-	// Parse CLI arguments.
-	options, err := config.FromArgs()
-	cli.HandleError(err)
-
-	logger := options.GetLogger()
-	logger.Info("using program arguments", "args", os.Args)
-	logger.Info("parsed CLI options", "options", options)
-
-	// Get input.
-	input, err := options.Source.Extract()
-	cli.HandleError(err)
-
-	// Parse code into syntax tree.
-	ast, err := saccharine.Parse(input)
-	cli.HandleError(err)
-	logger.Info("parsed syntax tree", "tree", ast)
-
-	// Compile expression to lambda calculus.
-	compiled := convert.SaccharineToLambda(ast)
-	logger.Info("compiled λ expression", "tree", compiled.String())
-
-	// Create reducer with the compiled expression.
-	runtime := normalorder.NewRuntime(compiled)
-
-	// If the user selected to track CPU performance, attach a profiler.
-	if options.Profile != "" {
-		plugins.NewPerformance(options.Profile, runtime)
+func Lambda() *cobra.Command {
+	cmd := &cobra.Command{
+		Use:   "lambda",
+		Short: "Lambda calculus interpreter",
+		Long:  "A lambda calculus interpreter supporting multiple representations.",
+		RunE: func(cmd *cobra.Command, _ []string) error {
+			return cmd.Help()
+		},
 	}
 
-	// If the user selected to produce a step-by-step explanation, attach an
-	// observer.
-	if options.Explanation {
-		plugins.NewExplanation(runtime)
-	}
+	cmd.AddCommand(LambdaConvert())
+	cmd.AddCommand(LambdaEngine())
+	cmd.AddCommand(LambdaReduce())
 
-	// If the user opted to track statistics, attach a tracker.
-	if options.Statistics {
-		plugins.NewStatistics(runtime)
-	}
-
-	// If the user selected for verbose debug logs, attach a reduction tracker.
-	if options.Verbose {
-		plugins.NewLogs(logger, runtime)
-	}
-
-	// Run reduction.
-	runtime.Run()
-
-	// Return the final reduced result.
-	result := runtime.Expression().String()
-	err = options.Destination.Write(result)
-	cli.HandleError(err)
+	return cmd
+}
+
+func main() {
+	lambda := Lambda()
+	if err := lambda.Execute(); err != nil {
+		os.Exit(1)
+	}
 }
diff --git a/cmd/lambda/lambda_convert.go b/cmd/lambda/lambda_convert.go
new file mode 100644
index 0000000..bd7ad9e
--- /dev/null
+++ b/cmd/lambda/lambda_convert.go
@@ -0,0 +1,94 @@
+package main
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/spf13/cobra"
+)
+
+// inferReprFromPath returns the repr type based on file extension.
+func inferReprFromPath(path string) (string, error) {
+	switch ext := strings.ToLower(filepath.Ext(path)); ext {
+	case ".lambda", ".lam", ".lc":
+		return "lambda", nil
+	case ".saccharine", ".sch":
+		return "saccharine", nil
+	default:
+		return "", fmt.Errorf("unknown file extension '%s'", ext)
+	}
+}
+
+func LambdaConvert() *cobra.Command {
+	var inputReprFlag, outputReprFlag string
+
+	cmd := &cobra.Command{
+		Use:          "convert <input-file> <output-file>",
+		Short:        "Convert between lambda calculus representations",
+		SilenceUsage: true,
+		RunE: func(cmd *cobra.Command, args []string) error {
+			if len(args) != 2 {
+				return cmd.Help()
+			}
+
+			var err error
+			inputPath, outputPath := args[0], args[1]
+
+			// Use flag if provided, otherwise infer from extension.
+			inputRepr := inputReprFlag
+			if inputRepr == "" {
+				if inputRepr, err = inferReprFromPath(inputPath); err != nil {
+					return fmt.Errorf("input file: %w", err)
+				}
+			}
+
+			outputRepr := outputReprFlag
+			if outputRepr == "" {
+				if outputRepr, err = inferReprFromPath(outputPath); err != nil {
+					return fmt.Errorf("output file: %w", err)
+				}
+			}
+
+			// Read input file.
+			input, err := os.ReadFile(inputPath)
+			if err != nil {
+				return fmt.Errorf("reading input file: %w", err)
+			}
+
+			r := GetRegistry()
+
+			// Parse input into syntax tree.
+			repr, err := r.Unmarshal(string(input), inputRepr)
+			if err != nil {
+				return fmt.Errorf("parsing input: %w", err)
+			}
+
+			// Convert to output repr if different.
+			result, err := r.ConvertTo(repr, outputRepr)
+			if err != nil {
+				return fmt.Errorf("converting %s to %s: %w", inputRepr, outputRepr, err)
+			}
+
+			// Marshal output.
+			output, err := r.Marshal(result)
+			if err != nil {
+				return fmt.Errorf("unmarshaling output: %w", err)
+			}
+
+			// Write output file.
+			err = os.WriteFile(outputPath, []byte(output), 0644)
+			if err != nil {
+				return fmt.Errorf("writing output file: %w", err)
+			}
+
+			return nil
+		},
+	}
+
+	cmd.Flags().StringVar(&inputReprFlag, "from", "", "Input representation (inferred from extension if unset)")
+	cmd.Flags().StringVar(&outputReprFlag, "to", "", "Output representation (inferred from extension if unset)")
+
+	return cmd
+}
diff --git a/cmd/lambda/lambda_engine.go b/cmd/lambda/lambda_engine.go
new file mode 100644
index 0000000..b8a2f35
--- /dev/null
+++ b/cmd/lambda/lambda_engine.go
@@ -0,0 +1,19 @@
+package main
+
+import (
+	"github.com/spf13/cobra"
+)
+
+func LambdaEngine() *cobra.Command {
+	cmd := &cobra.Command{
+		Use:   "engine",
+		Short: "Information about available engines",
+		RunE: func(cmd *cobra.Command, args []string) error {
+			return cmd.Help()
+		},
+	}
+
+	cmd.AddCommand(LambdaEngineList())
+
+	return cmd
+}
diff --git a/cmd/lambda/lambda_engine_list.go b/cmd/lambda/lambda_engine_list.go
new file mode 100644
index 0000000..338098d
--- /dev/null
+++ b/cmd/lambda/lambda_engine_list.go
@@ -0,0 +1,26 @@
+package main
+
+import (
+	"fmt"
+
+	"github.com/spf13/cobra"
+)
+
+func LambdaEngineList() *cobra.Command {
+	cmd := &cobra.Command{
+		Use:     "list",
+		Aliases: []string{"ls"},
+		Short:   "List available engines",
+		RunE: func(cmd *cobra.Command, args []string) error {
+			r := GetRegistry()
+
+			for engine := range r.ListEngines() {
+				fmt.Println(engine.Name())
+			}
+
+			return nil
+		},
+	}
+
+	return cmd
+}
diff --git a/cmd/lambda/lambda_reduce.go b/cmd/lambda/lambda_reduce.go
new file mode 100644
index 0000000..a6f1039
--- /dev/null
+++ b/cmd/lambda/lambda_reduce.go
@@ -0,0 +1,109 @@
+package main
+
+import (
+	"fmt"
+
+	"github.com/spf13/cobra"
+
+	"git.maximhutz.com/max/lambda/internal/cli"
+	"git.maximhutz.com/max/lambda/internal/config"
+)
+
+func LambdaReduce() *cobra.Command {
+	var inputReprFlag string
+	var engineFlag string
+
+	cmd := &cobra.Command{
+		Use:          "reduce <input-file>",
+		Short:        "Reduce a lambda calculus expression",
+		SilenceUsage: true,
+		Aliases:      []string{"run"},
+		RunE: func(cmd *cobra.Command, args []string) error {
+			var err error
+			if len(args) != 1 {
+				return cmd.Help()
+			}
+
+			inputPath := args[0]
+
+			// Get input source.
+			var source config.Source
+			if inputPath == "-" {
+				source = config.StdinSource{}
+			} else {
+				source = config.FileSource{Path: inputPath}
+			}
+
+			destination := config.StdoutDestination{}
+
+			r := GetRegistry()
+
+			// Get input.
+			input, err := source.Extract()
+			if err != nil {
+				return err
+			}
+
+			// Use flag if provided, otherwise infer from extension.
+			inputRepr := inputReprFlag
+			if inputRepr == "" {
+				if inputRepr, err = inferReprFromPath(inputPath); err != nil {
+					return fmt.Errorf("input file: %w", err)
+				}
+			}
+
+			// Find engine.
+			var engine cli.Engine
+			if engineFlag == "" {
+				if engine, err = r.GetDefaultEngine(inputRepr); err != nil {
+					return err
+				}
+			} else {
+				if engine, err = r.GetEngine(engineFlag); err != nil {
+					return err
+				}
+			}
+
+			// Parse code into syntax tree.
+			repr, err := r.Unmarshal(input, inputRepr)
+			if err != nil {
+				return err
+			}
+
+			// Compile expression to lambda calculus.
+			compiled, err := r.ConvertTo(repr, "lambda")
+			if err != nil {
+				return err
+			}
+
+			// Create process.
+			process := engine.Load()
+			err = process.Set(compiled)
+			if err != nil {
+				return err
+			}
+
+			// Run reduction.
+			for process.Step(1) {
+			}
+
+			// Return the final reduced result.
+			result, err := process.Get()
+			if err != nil {
+				return err
+			}
+
+			output, err := r.Marshal(result)
+			if err != nil {
+				return err
+			}
+
+			return destination.Write(output)
+		},
+	}
+
+	cmd.Flags().StringVar(&inputReprFlag, "from", "", "Input representation (inferred from extension if unset)")
+	cmd.Flags().StringVarP(&engineFlag, "engine", "e", "", "Reduction engine (inferred from '--input' if unset)")
+
+	return cmd
+}
diff --git a/cmd/lambda/lambda_test.go b/cmd/lambda/lambda_test.go
deleted file mode 100644
index cbca46b..0000000
--- a/cmd/lambda/lambda_test.go
+++ /dev/null
@@ -1,85 +0,0 @@
-package main
-
-import (
-	"os"
-	"path/filepath"
-	"strings"
-	"testing"
-
-	"git.maximhutz.com/max/lambda/pkg/convert"
-	"git.maximhutz.com/max/lambda/pkg/normalorder"
-	"git.maximhutz.com/max/lambda/pkg/saccharine"
-	"github.com/stretchr/testify/assert"
-)
-
-// Helper function to run a single sample through the lambda runtime.
-func runSample(samplePath string) (string, error) {
-	// Read the sample file.
-	input, err := os.ReadFile(samplePath)
-	if err != nil {
-		return "", err
-	}
-
-	// Parse code into syntax tree.
-	ast, err := saccharine.Parse(string(input))
-	if err != nil {
-		return "", err
-	}
-
-	// Compile expression to lambda calculus.
-	compiled := convert.SaccharineToLambda(ast)
-
-	// Create and run the reducer.
-	reducer := normalorder.NewRuntime(compiled)
-	reducer.Run()
-
-	return reducer.Expression().String() + "\n", nil
-}
-
-// Test that all samples produce expected output.
-func TestSamplesValidity(t *testing.T) {
-	// Discover all .test files in the tests directory.
-	testFiles, err := filepath.Glob("../../tests/*.test")
-	assert.NoError(t, err, "Failed to read tests directory.")
-	assert.NotEmpty(t, testFiles, "No '*.test' files found in directory.")
-
-	for _, testPath := range testFiles {
-		// Build expected file path.
-		expectedPath := strings.TrimSuffix(testPath, filepath.Ext(testPath)) + ".expected"
-
-		name := strings.TrimSuffix(filepath.Base(testPath), filepath.Ext(testPath))
-
-		t.Run(name, func(t *testing.T) {
-			// Run the sample and capture output.
-			actual, err := runSample(testPath)
-			assert.NoError(t, err, "Failed to run sample.")
-
-			// Read expected output.
-			expectedBytes, err := os.ReadFile(expectedPath)
-			assert.NoError(t, err, "Failed to read expected output.")
-			expected := string(expectedBytes)
-
-			// Compare outputs.
-			assert.Equal(t, expected, actual, "Output does not match expected.")
-		})
-	}
-}
-
-// Benchmark all samples using sub-benchmarks.
-func BenchmarkSamples(b *testing.B) {
-	// Discover all .test files in the tests directory.
-	testFiles, err := filepath.Glob("../../tests/*.test")
-	assert.NoError(b, err, "Failed to read tests directory.")
-	assert.NotEmpty(b, testFiles, "No '*.test' files found in directory.")
-
-	for _, path := range testFiles {
-		name := strings.TrimSuffix(filepath.Base(path), filepath.Ext(path))
-
-		b.Run(name, func(b *testing.B) {
-			for b.Loop() {
-				_, err := runSample(path)
-				assert.NoError(b, err, "Failed to run sample.")
-			}
-		})
-	}
-}
diff --git a/cmd/lambda/registry.go b/cmd/lambda/registry.go
new file mode 100644
index 0000000..5603971
--- /dev/null
+++ b/cmd/lambda/registry.go
@@ -0,0 +1,26 @@
+package main
+
+import (
+	"git.maximhutz.com/max/lambda/internal/cli"
+	"git.maximhutz.com/max/lambda/internal/registry"
+	"git.maximhutz.com/max/lambda/pkg/convert"
+	"git.maximhutz.com/max/lambda/pkg/engine/normalorder"
+	"git.maximhutz.com/max/lambda/pkg/lambda"
+	"git.maximhutz.com/max/lambda/pkg/saccharine"
+)
+
+func GetRegistry() *registry.Registry {
+	r := registry.New()
+
+	// Codecs
+	r.MustAddConversions(cli.ConvertCodec(convert.Saccharine2Lambda{}, "saccharine", "lambda")...)
+
+	// Engines
+	r.MustAddEngine(cli.ConvertEngine(normalorder.Engine{}, "normalorder", "lambda"))
+
+	// Marshalers
+	r.MustAddMarshaler(cli.ConvertMarshaler(lambda.Marshaler{}, "lambda"))
+	r.MustAddMarshaler(cli.ConvertMarshaler(saccharine.Marshaler{}, "saccharine"))
+
+	return r
+}
diff --git a/go.mod b/go.mod
index 241b0b4..3633877 100644
--- a/go.mod
+++ b/go.mod
@@ -2,10 +2,9 @@ module git.maximhutz.com/max/lambda
 
 go 1.25.5
 
-require github.com/stretchr/testify v1.11.1
+require github.com/spf13/cobra v1.10.2
 
 require (
-	github.com/davecgh/go-spew v1.1.1 // indirect
-	github.com/pmezard/go-difflib v1.0.0 // indirect
-	gopkg.in/yaml.v3 v3.0.1 // indirect
+	github.com/inconshreveable/mousetrap v1.1.0 // indirect
+	github.com/spf13/pflag v1.0.10 // indirect
 )
diff --git a/go.sum b/go.sum
index cc8b3f4..ef5d78d 100644
--- a/go.sum
+++ b/go.sum
@@ -1,9 +1,11 @@
-github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
-github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
-github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
-github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
-github.com/stretchr/testify v1.11.1 h1:7s2iGBzp5EwR7/aIZr8ao5+dra3wiQyKjjFuvgVKu7U=
-github.com/stretchr/testify v1.11.1/go.mod h1:wZwfW3scLgRK+23gO65QZefKpKQRnfz6sD981Nm4B6U=
+github.com/cpuguy83/go-md2man/v2 v2.0.6/go.mod h1:oOW0eioCTA6cOiMLiUPZOpcVxMig6NIQQ7OS05n1F4g=
+github.com/inconshreveable/mousetrap v1.1.0 h1:wN+x4NVGpMsO7ErUn/mUI3vEoE6Jt13X2s0bqwp9tc8=
+github.com/inconshreveable/mousetrap v1.1.0/go.mod h1:vpF70FUmC8bwa3OWnCshd2FqLfsEA9PFc4w1p2J65bw=
+github.com/russross/blackfriday/v2 v2.1.0/go.mod h1:+Rmxgy9KzJVeS9/2gXHxylqXiyQDYRxCVz55jmeOWTM=
+github.com/spf13/cobra v1.10.2 h1:DMTTonx5m65Ic0GOoRY2c16WCbHxOOw6xxezuLaBpcU=
+github.com/spf13/cobra v1.10.2/go.mod h1:7C1pvHqHw5A4vrJfjNwvOdzYu0Gml16OCs2GRiTUUS4=
+github.com/spf13/pflag v1.0.9/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An2Bg=
+github.com/spf13/pflag v1.0.10 h1:4EBh2KAYBwaONj6b2Ye1GiHfwjqyROoF4RwYO+vPwFk=
+github.com/spf13/pflag v1.0.10/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An2Bg=
+go.yaml.in/yaml/v3 v3.0.4/go.mod h1:DhzuOOF2ATzADvBadXxruRBLzYTpT36CKvDb3+aBEFg=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
-gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
-gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
diff --git a/internal/cli/conversion.go b/internal/cli/conversion.go
new file mode 100644
index 0000000..d198f8e
--- /dev/null
+++ b/internal/cli/conversion.go
@@ -0,0 +1,67 @@
+package cli
+
+import (
+	"fmt"
+
+	"git.maximhutz.com/max/lambda/pkg/codec"
+)
+
+type Conversion interface {
+	InType() string
+	OutType() string
+
+	Run(Repr) (Repr, error)
+}
+
+type forwardCodec[T, U any] struct {
+	codec           codec.Codec[T, U]
+	inType, outType string
+}
+
+func (c forwardCodec[T, U]) Run(r Repr) (Repr, error) {
+	t, ok := r.Data().(T)
+	if !ok {
+		return nil, fmt.Errorf("could not parse '%v' as '%s'", t, c.inType)
+	}
+
+	u, err := c.codec.Encode(t)
+	if err != nil {
+		return nil, err
+	}
+
+	return NewRepr(c.outType, u), nil
+}
+
+func (c forwardCodec[T, U]) InType() string { return c.inType }
+
+func (c forwardCodec[T, U]) OutType() string { return c.outType }
+
+type backwardCodec[T, U any] struct {
+	codec           codec.Codec[T, U]
+	inType, outType string
+}
+
+func (c backwardCodec[T, U]) Run(r Repr) (Repr, error) {
+	u, ok := r.Data().(U)
+	if !ok {
+		return nil, fmt.Errorf("could not parse '%v' as '%s'", r, c.outType)
+	}
+
+	t, err := c.codec.Decode(u)
+	if err != nil {
+		return nil, err
+	}
+
+	return NewRepr(c.inType, t), nil
+}
+
+func (c backwardCodec[T, U]) InType() string { return c.outType }
+
+func (c backwardCodec[T, U]) OutType() string { return c.inType }
+
+func ConvertCodec[T, U any](e codec.Codec[T, U], inType, outType string) []Conversion {
+	return []Conversion{
+		forwardCodec[T, U]{e, inType, outType},
+		backwardCodec[T, U]{e, inType, outType},
+	}
+}
diff --git a/internal/cli/engine.go b/internal/cli/engine.go
new file mode 100644
index 0000000..15f1618
--- /dev/null
+++ b/internal/cli/engine.go
@@ -0,0 +1,27 @@
+package cli
+
+import "git.maximhutz.com/max/lambda/pkg/engine"
+
+type Engine interface {
+	Load() Process
+	Name() string
+	InType() string
+}
+
+type convertedEngine[T any] struct {
+	engine engine.Engine[T]
+	name   string
+	inType string
+}
+
+func (e convertedEngine[T]) InType() string { return e.inType }
+
+func (e convertedEngine[T]) Name() string { return e.name }
+
+func (e convertedEngine[T]) Load() Process {
+	return convertedProcess[T]{e.engine.Load(), e.inType}
+}
+
+func ConvertEngine[T any](e engine.Engine[T], name, inType string) Engine {
+	return &convertedEngine[T]{e, name, inType}
+}
diff --git a/internal/cli/exit.go b/internal/cli/exit.go
deleted file mode 100644
index 371ecee..0000000
--- a/internal/cli/exit.go
+++ /dev/null
@@ -1,18 +0,0 @@
-// Package "cli" provides miscellaneous helper functions.
-package cli
-
-import (
-	"fmt"
-	"os"
-)
-
-// A helper function to handle errors in the program. If it is given an error,
-// the program will exist, and print the error.
-func HandleError(err error) {
-	if err == nil {
-		return
-	}
-
-	fmt.Fprintln(os.Stderr, "ERROR:", err)
-	os.Exit(1)
-}
diff --git a/internal/cli/marshaler.go b/internal/cli/marshaler.go
new file mode 100644
index 0000000..a212417
--- /dev/null
+++ b/internal/cli/marshaler.go
@@ -0,0 +1,45 @@
+package cli
+
+import (
+	"fmt"
+	"reflect"
+
+	"git.maximhutz.com/max/lambda/pkg/codec"
+)
+
+type Marshaler interface {
+	codec.Marshaler[Repr]
+
+	InType() string
+}
+
+type convertedMarshaler[T any] struct {
+	codec  codec.Marshaler[T]
+	inType string
+}
+
+func (c convertedMarshaler[T]) Decode(s string) (Repr, error) {
+	t, err := c.codec.Decode(s)
+	if err != nil {
+		return nil, err
+	}
+
+	return NewRepr(c.inType, t), nil
+}
+
+func (c convertedMarshaler[T]) Encode(r Repr) (string, error) {
+	t, ok := r.Data().(T)
+	if !ok {
+		dataType := reflect.TypeOf(r.Data())
+		allowedType := reflect.TypeFor[T]()
+		return "", fmt.Errorf("marshaler for '%s' cannot parse '%s'", allowedType, dataType)
+	}
+
+	return c.codec.Encode(t)
+}
+
+func (c convertedMarshaler[T]) InType() string { return c.inType }
+
+func ConvertMarshaler[T any](e codec.Marshaler[T], inType string) Marshaler {
+	return convertedMarshaler[T]{e, inType}
+}
diff --git a/internal/cli/process.go b/internal/cli/process.go
new file mode 100644
index 0000000..f3445f7
--- /dev/null
+++ b/internal/cli/process.go
@@ -0,0 +1,41 @@
+package cli
+
+import (
+	"fmt"
+
+	"git.maximhutz.com/max/lambda/pkg/engine"
+)
+
+type Process interface {
+	engine.Process[Repr]
+
+	InType() string
+}
+
+type convertedProcess[T any] struct {
+	process engine.Process[T]
+	inType  string
+}
+
+func (e convertedProcess[T]) InType() string { return e.inType }
+
+func (b convertedProcess[T]) Get() (Repr, error) {
+	s, err := b.process.Get()
+	if err != nil {
+		return nil, err
+	}
+
+	return NewRepr(b.inType, s), nil
+}
+
+func (b convertedProcess[T]) Set(r Repr) error {
+	if t, ok := r.Data().(T); ok {
+		return b.process.Set(t)
+	}
+
+	return fmt.Errorf("Incorrent format '%s' for engine '%s'.", r.Id(), b.inType)
+}
+
+func (b convertedProcess[T]) Step(i int) bool {
+	return b.process.Step(i)
+}
diff --git a/internal/cli/repr.go b/internal/cli/repr.go
new file mode 100644
index 0000000..c684639
--- /dev/null
+++ b/internal/cli/repr.go
@@ -0,0 +1,21 @@
+package cli
+
+type Repr interface {
+	// Id returns to name of the objects underlying representation. If is
+	// assumed that if two Repr objects have the same Id(), they share the same
+	// representation.
+	Id() string
+
+	Data() any
+}
+
+type baseRepr struct {
+	id   string
+	data any
+}
+
+func (r baseRepr) Id() string { return r.id }
+
+func (r baseRepr) Data() any { return r.data }
+
+func NewRepr(id string, data any) Repr { return baseRepr{id, data} }
diff --git a/internal/plugins/debug.go b/internal/plugins/debug.go
deleted file mode 100644
index 82cd1b8..0000000
--- a/internal/plugins/debug.go
+++ /dev/null
@@ -1,23 +0,0 @@
-package plugins
-
-import (
-	"log/slog"
-
-	"git.maximhutz.com/max/lambda/pkg/runtime"
-)
-
-type Logs struct {
-	logger  *slog.Logger
-	reducer runtime.Runtime
-}
-
-func NewLogs(logger *slog.Logger, r runtime.Runtime) *Logs {
-	plugin := &Logs{logger, r}
-	r.On(runtime.StepEvent, plugin.Step)
-
-	return plugin
-}
-
-func (t *Logs) Step() {
-	t.logger.Info("reduction", "tree", t.reducer.Expression().String())
-}
diff --git a/internal/plugins/explanation.go b/internal/plugins/explanation.go
deleted file mode 100644
index ba92a99..0000000
--- a/internal/plugins/explanation.go
+++ /dev/null
@@ -1,31 +0,0 @@
-// Package "explanation" provides an observer to gather the reasoning during the
-// reduction, and present a thorough explanation to the user for each step.
-package plugins
-
-import (
-	"fmt"
-
-	"git.maximhutz.com/max/lambda/pkg/runtime"
-)
-
-// Track the reductions made by a reduction process.
-type Explanation struct {
-	reducer runtime.Runtime
-}
-
-// Attaches a new explanation tracker to a reducer.
-func NewExplanation(r runtime.Runtime) *Explanation {
-	plugin := &Explanation{reducer: r}
-	r.On(runtime.StartEvent, plugin.Start)
-	r.On(runtime.StepEvent, plugin.Step)
-
-	return plugin
-}
-
-func (t *Explanation) Start() {
-	fmt.Println(t.reducer.Expression().String())
-}
-
-func (t *Explanation) Step() {
-	fmt.Println("    =", t.reducer.Expression().String())
-}
diff --git a/internal/plugins/performance.go b/internal/plugins/performance.go
deleted file mode 100644
index eac3f20..0000000
--- a/internal/plugins/performance.go
+++ /dev/null
@@ -1,59 +0,0 @@
-// Package "performance" provides a tracker to observer CPU performance during
-// execution.
-package plugins
-
-import (
-	"os"
-	"path/filepath"
-	"runtime/pprof"
-
-	"git.maximhutz.com/max/lambda/pkg/runtime"
-)
-
-// Observes a reduction process, and publishes a CPU performance profile on
-// completion.
-type Performance struct {
-	File        string
-	filePointer *os.File
-	Error       error
-}
-
-// Create a performance tracker that outputs a profile to "file".
-func NewPerformance(file string, process runtime.Runtime) *Performance {
-	plugin := &Performance{File: file}
-	process.On(runtime.StartEvent, plugin.Start)
-	process.On(runtime.StopEvent, plugin.Stop)
-
-	return plugin
-}
-
-// Begin profiling.
-func (t *Performance) Start() {
-	var absPath string
-
-	absPath, t.Error = filepath.Abs(t.File)
-	if t.Error != nil {
-		return
-	}
-
-	t.Error = os.MkdirAll(filepath.Dir(absPath), 0777)
-	if t.Error != nil {
-		return
-	}
-
-	t.filePointer, t.Error = os.Create(absPath)
-	if t.Error != nil {
-		return
-	}
-
-	t.Error = pprof.StartCPUProfile(t.filePointer)
-	if t.Error != nil {
-		return
-	}
-}
-
-// Stop profiling.
-func (t *Performance) Stop() {
-	pprof.StopCPUProfile()
-	t.filePointer.Close()
-}
diff --git a/internal/plugins/statistics.go b/internal/plugins/statistics.go
deleted file mode 100644
index 2845738..0000000
--- a/internal/plugins/statistics.go
+++ /dev/null
@@ -1,44 +0,0 @@
-package plugins
-
-import (
-	"fmt"
-	"os"
-	"time"
-
-	"git.maximhutz.com/max/lambda/internal/statistics"
-	"git.maximhutz.com/max/lambda/pkg/runtime"
-)
-
-// An observer, to track reduction performance.
-type Statistics struct {
-	start time.Time
-	steps uint64
-}
-
-// Create a new reduction performance Statistics.
-func NewStatistics(r runtime.Runtime) *Statistics {
-	plugin := &Statistics{}
-	r.On(runtime.StartEvent, plugin.Start)
-	r.On(runtime.StepEvent, plugin.Step)
-	r.On(runtime.StopEvent, plugin.Stop)
-
-	return plugin
-}
-
-func (t *Statistics) Start() {
-	t.start = time.Now()
-	t.steps = 0
-}
-
-func (t *Statistics) Step() {
-	t.steps++
-}
-
-func (t *Statistics) Stop() {
-	results := statistics.Results{
-		StepsTaken:  t.steps,
-		TimeElapsed: uint64(time.Since(t.start).Milliseconds()),
-	}
-
-	fmt.Fprint(os.Stderr, results.String())
-}
diff --git a/internal/registry/converter.go b/internal/registry/converter.go
new file mode 100644
index 0000000..3c29199
--- /dev/null
+++ b/internal/registry/converter.go
@@ -0,0 +1,27 @@
+package registry
+
+import (
+	"git.maximhutz.com/max/lambda/internal/cli"
+)
+
+type Converter struct {
+	data map[string][]cli.Conversion
+}
+
+func NewConverter() *Converter {
+	return &Converter{data: map[string][]cli.Conversion{}}
+}
+
+func (g *Converter) Add(c cli.Conversion) {
+	conversionsFromIn, ok := g.data[c.InType()]
+	if !ok {
+		conversionsFromIn = []cli.Conversion{}
+	}
+
+	conversionsFromIn = append(conversionsFromIn, c)
+	g.data[c.InType()] = conversionsFromIn
+}
+
+func (g *Converter) ConversionsFrom(t string) []cli.Conversion {
+	return g.data[t]
+}
diff --git a/internal/registry/registry.go b/internal/registry/registry.go
new file mode 100644
index 0000000..705b405
--- /dev/null
+++ b/internal/registry/registry.go
@@ -0,0 +1,173 @@
+package registry
+
+import (
+	"fmt"
+	"iter"
+	"maps"
+
+	"git.maximhutz.com/max/lambda/internal/cli"
+)
+
+type Registry struct {
+	marshalers map[string]cli.Marshaler
+	converter  *Converter
+	engines    map[string]cli.Engine
+}
+
+func New() *Registry {
+	return &Registry{
+		marshalers: map[string]cli.Marshaler{},
+		converter:  NewConverter(),
+		engines:    map[string]cli.Engine{},
+	}
+}
+
+func (r *Registry) AddConversions(conversions ...cli.Conversion) error {
+	for _, conversion := range conversions {
+		r.converter.Add(conversion)
+	}
+
+	return nil
+}
+func (r *Registry) MustAddConversions(conversions ...cli.Conversion) {
+	if err := r.AddConversions(conversions...); err != nil {
+		panic(err)
+	}
+}
+
+func (r *Registry) AddMarshaler(c cli.Marshaler) error {
+	if _, ok := r.marshalers[c.InType()]; ok {
+		return fmt.Errorf("marshaler for '%s' already registered", c.InType())
+	}
+
+	r.marshalers[c.InType()] = c
+	return nil
+}
+
+func (r *Registry) MustAddMarshaler(c cli.Marshaler) {
+	if err := r.AddMarshaler(c); err != nil {
+		panic(err)
+	}
+}
+
+func (r *Registry) AddEngine(e cli.Engine) error {
+	if _, ok := r.engines[e.Name()]; ok {
+		return fmt.Errorf("engine '%s' already registered", e.Name())
+	}
+
+	r.engines[e.Name()] = e
+	return nil
+}
+
+func (r *Registry) MustAddEngine(e cli.Engine) {
+	if err := r.AddEngine(e); err != nil {
+		panic(err)
+	}
+}
+
+func (r Registry) GetEngine(name string) (cli.Engine, error) {
+	e, ok := r.engines[name]
+	if !ok {
+		return nil, fmt.Errorf("engine '%s' not found", name)
+	}
+
+	return e, nil
+}
+
+func (r Registry) ListEngines() iter.Seq[cli.Engine] {
+	return maps.Values(r.engines)
+}
+
+func (r *Registry) GetDefaultEngine(id string) (cli.Engine, error) {
+	for _, engine := range r.engines {
+		if engine.InType() == id {
+			return engine, nil
+		}
+	}
+
+	return nil, fmt.Errorf("no engine for '%s'", id)
+}
+
+func (r *Registry) ConvertTo(repr cli.Repr, outType string) (cli.Repr, error) {
+	path, err := r.ConversionPath(repr.Id(), outType)
+	if err != nil {
+		return nil, err
+	}
+
+	result := repr
+	for _, conversion := range path {
+		result, err = conversion.Run(result)
+		if err != nil {
+			return nil, fmt.Errorf("converting '%s' to '%s': %w", conversion.InType(), conversion.OutType(), err)
+		}
+	}
+
+	return result, err
+}
+
+func (r *Registry) Marshal(repr cli.Repr) (string, error) {
+	m, ok := r.marshalers[repr.Id()]
+	if !ok {
+		return "", fmt.Errorf("no marshaler for '%s'", repr.Id())
+	}
+
+	return m.Encode(repr)
+}
+
+func (r *Registry) Unmarshal(s string, outType string) (cli.Repr, error) {
+	m, ok := r.marshalers[outType]
+	if !ok {
+		return nil, fmt.Errorf("no marshaler for '%s'", outType)
+	}
+
+	return m.Decode(s)
+}
+
+func reverse[T any](list []T) []T {
+	if list == nil {
+		return list
+	}
+
+	reversed := []T{}
+
+	for i := len(list) - 1; i >= 0; i-- {
+		reversed = append(reversed, list[i])
+	}
+
+	return reversed
+}
+
+func (r *Registry) ConversionPath(from, to string) ([]cli.Conversion, error) {
+	backtrack := map[string]cli.Conversion{}
+	iteration := []string{from}
+	for len(iteration) > 0 {
+		nextIteration := []string{}
+
+		for _, item := range iteration {
+			for _, conversion := range r.converter.ConversionsFrom(item) {
+				if _, ok := backtrack[conversion.OutType()]; ok {
+					continue
+				}
+
+				nextIteration = append(nextIteration, conversion.OutType())
+				backtrack[conversion.OutType()] = conversion
+			}
+		}
+
+		iteration = nextIteration
+	}
+
+	reversedPath := []cli.Conversion{}
+	current := to
+	for current != from {
+		conversion, ok := backtrack[current]
+		if !ok {
+			return nil, fmt.Errorf("no valid conversion from '%s' to '%s'", from, to)
+		}
+
+		reversedPath = append(reversedPath, conversion)
+		current = conversion.InType()
+	}
+
+	return reverse(reversedPath), nil
+}
diff --git a/internal/statistics/statistics.go b/internal/statistics/statistics.go
deleted file mode 100644
index 87cf386..0000000
--- a/internal/statistics/statistics.go
+++ /dev/null
@@ -1,28 +0,0 @@
-// Package "statistics" provides a way to observer reduction speed during
-// execution.
-package statistics
-
-import (
-	"fmt"
-	"strings"
-)
-
-// Statistics for a specific reduction.
-type Results struct {
-	StepsTaken  uint64 // Number of steps taken during execution.
-	TimeElapsed uint64 // The time (ms) taken for execution to complete.
-}
-
-// Returns the average number of operations per second of the execution.
-func (r Results) OpsPerSecond() float32 {
-	return float32(r.StepsTaken) / (float32(r.TimeElapsed) / 1000)
-}
-
-// Format the results as a string.
-func (r Results) String() string {
-	builder := strings.Builder{}
-	fmt.Fprintln(&builder, "Time Spent:", r.TimeElapsed, "ms")
-	fmt.Fprintln(&builder, "Steps:", r.StepsTaken)
-	fmt.Fprintln(&builder, "Speed:", r.OpsPerSecond(), "ops")
-	return builder.String()
-}
diff --git a/pkg/codec/codec.go b/pkg/codec/codec.go
new file mode 100644
index 0000000..3132409
--- /dev/null
+++ b/pkg/codec/codec.go
@@ -0,0 +1,8 @@
+package codec
+
+type Codec[T, U any] interface {
+	Encode(T) (U, error)
+	Decode(U) (T, error)
+}
+
+type Marshaler[T any] = Codec[T, string]
diff --git a/pkg/convert/saccharine_to_lambda.go b/pkg/convert/saccharine_to_lambda.go
index 8bab6b8..e19e77c 100644
--- a/pkg/convert/saccharine_to_lambda.go
+++ b/pkg/convert/saccharine_to_lambda.go
@@ -3,16 +3,17 @@ package convert
 import (
 	"fmt"
 
+	"git.maximhutz.com/max/lambda/pkg/codec"
 	"git.maximhutz.com/max/lambda/pkg/lambda"
 	"git.maximhutz.com/max/lambda/pkg/saccharine"
 )
 
-func convertAtom(n *saccharine.Atom) lambda.Expression {
+func encodeAtom(n *saccharine.Atom) lambda.Expression {
 	return lambda.NewVariable(n.Name)
 }
 
-func convertAbstraction(n *saccharine.Abstraction) lambda.Expression {
-	result := SaccharineToLambda(n.Body)
+func encodeAbstraction(n *saccharine.Abstraction) lambda.Expression {
+	result := encodeExpression(n.Body)
 
 	parameters := n.Parameters
 
@@ -31,13 +32,13 @@ func convertAbstraction(n *saccharine.Abstraction) lambda.Expression {
 	return result
 }
 
-func convertApplication(n *saccharine.Application) lambda.Expression {
-	result := SaccharineToLambda(n.Abstraction)
+func encodeApplication(n *saccharine.Application) lambda.Expression {
+	result := encodeExpression(n.Abstraction)
 
 	arguments := []lambda.Expression{}
 	for _, argument := range n.Arguments {
-		convertedArgument := SaccharineToLambda(argument)
-		arguments = append(arguments, convertedArgument)
+		encodeedArgument := encodeExpression(argument)
+		arguments = append(arguments, encodeedArgument)
 	}
 
 	for _, argument := range arguments {
@@ -51,9 +52,9 @@ func reduceLet(s *saccharine.LetStatement, e lambda.Expression) lambda.Expressio
 	var value lambda.Expression
 
 	if len(s.Parameters) == 0 {
-		value = SaccharineToLambda(s.Body)
+		value = encodeExpression(s.Body)
 	} else {
-		value = convertAbstraction(saccharine.NewAbstraction(s.Parameters, s.Body))
+		value = encodeAbstraction(saccharine.NewAbstraction(s.Parameters, s.Body))
 	}
 
 	return lambda.NewApplication(
@@ -67,7 +68,7 @@ func reduceDeclare(s *saccharine.DeclareStatement, e lambda.Expression) lambda.E
 
 	return lambda.NewApplication(
 		lambda.NewAbstraction(freshVar, e),
-		SaccharineToLambda(s.Value),
+		encodeExpression(s.Value),
 	)
 }
 
@@ -82,8 +83,8 @@ func reduceStatement(s saccharine.Statement, e lambda.Expression) lambda.Express
 	}
 }
 
-func convertClause(n *saccharine.Clause) lambda.Expression {
-	result := SaccharineToLambda(n.Returns)
+func encodeClause(n *saccharine.Clause) lambda.Expression {
+	result := encodeExpression(n.Returns)
 
 	for i := len(n.Statements) - 1; i >= 0; i-- {
 		result = reduceStatement(n.Statements[i], result)
@@ -92,17 +93,46 @@ func convertClause(n *saccharine.Clause) lambda.Expression {
 	return result
 }
 
-func SaccharineToLambda(n saccharine.Expression) lambda.Expression {
-	switch n := n.(type) {
+func encodeExpression(s saccharine.Expression) lambda.Expression {
+	switch s := s.(type) {
 	case *saccharine.Atom:
-		return convertAtom(n)
+		return encodeAtom(s)
 	case *saccharine.Abstraction:
-		return convertAbstraction(n)
+		return encodeAbstraction(s)
 	case *saccharine.Application:
-		return convertApplication(n)
+		return encodeApplication(s)
 	case *saccharine.Clause:
-		return convertClause(n)
+		return encodeClause(s)
 	default:
-		panic(fmt.Errorf("unknown expression type: %T", n))
+		panic(fmt.Errorf("unknown expression type: %T", s))
 	}
 }
+
+func decodeExression(l lambda.Expression) saccharine.Expression {
+	switch l := l.(type) {
+	case lambda.Variable:
+		return saccharine.NewAtom(l.Name())
+	case lambda.Abstraction:
+		return saccharine.NewAbstraction(
+			[]string{l.Parameter()},
+			decodeExression(l.Body()))
+	case lambda.Application:
+		return saccharine.NewApplication(
+			decodeExression(l.Abstraction()),
+			[]saccharine.Expression{decodeExression(l.Argument())})
+	default:
+		panic(fmt.Errorf("unknown expression type: %T", l))
+	}
+}
+
+type Saccharine2Lambda struct{}
+
+func (c Saccharine2Lambda) Decode(l lambda.Expression) (saccharine.Expression, error) {
+	return decodeExression(l), nil
+}
+
+func (c Saccharine2Lambda) Encode(s saccharine.Expression) (lambda.Expression, error) {
+	return encodeExpression(s), nil
+}
+
+var _ codec.Codec[saccharine.Expression, lambda.Expression] = (*Saccharine2Lambda)(nil)
diff --git a/pkg/engine/engine.go b/pkg/engine/engine.go
new file mode 100644
index 0000000..0fcebf3
--- /dev/null
+++ b/pkg/engine/engine.go
@@ -0,0 +1,11 @@
+package engine
+
+type Engine[T any] interface {
+	Load() Process[T]
+}
+
+type Process[T any] interface {
+	Get() (T, error)
+	Set(T) error
+	Step(int) bool
+}
diff --git a/pkg/engine/normalorder/engine.go b/pkg/engine/normalorder/engine.go
new file mode 100644
index 0000000..fcdd362
--- /dev/null
+++ b/pkg/engine/normalorder/engine.go
@@ -0,0 +1,42 @@
+package normalorder
+
+import (
+	"git.maximhutz.com/max/lambda/pkg/engine"
+	"git.maximhutz.com/max/lambda/pkg/lambda"
+)
+
+type Process struct {
+	expr lambda.Expression
+}
+
+func (e Process) Get() (lambda.Expression, error) {
+	return e.expr, nil
+}
+
+func (e *Process) Set(l lambda.Expression) error {
+	e.expr = l
+	return nil
+}
+
+func (e *Process) Step(i int) bool {
+	for range i {
+		next, reduced := ReduceOnce(e.expr)
+		if !reduced {
+			return false
+		}
+
+		e.expr = next
+	}
+
+	return true
+}
+
+type Engine struct {
+}
+
+func (e Engine) Load() engine.Process[lambda.Expression] {
+	return &Process{}
+}
+
+var _ engine.Process[lambda.Expression] = (*Process)(nil)
+var _ engine.Engine[lambda.Expression] = (*Engine)(nil)
diff --git a/pkg/normalorder/reduce_once.go b/pkg/engine/normalorder/reduce_one.go
similarity index 100%
rename from pkg/normalorder/reduce_once.go
rename to pkg/engine/normalorder/reduce_one.go
diff --git a/pkg/expr/expr.go b/pkg/expr/expr.go
deleted file mode 100644
index b45a515..0000000
--- a/pkg/expr/expr.go
+++ /dev/null
@@ -1,15 +0,0 @@
-// Package expr provides the abstract Expression interface for all evaluatable
-// expression types in the lambda runtime.
-package expr
-
-import (
-	"fmt"
-)
-
-// Expression is the base interface for all evaluatable expression types.
-// Different evaluation modes (lambda calculus, SKI combinators, typed lambda
-// calculus, etc.) implement this interface with their own concrete types.
-type Expression interface {
-	// The expression should have a human-readable representation.
-	fmt.Stringer
-}
diff --git a/pkg/lambda/expression.go b/pkg/lambda/expression.go
index dd28fe6..b979a09 100644
--- a/pkg/lambda/expression.go
+++ b/pkg/lambda/expression.go
@@ -1,14 +1,15 @@
 package lambda
 
 import (
-	"git.maximhutz.com/max/lambda/pkg/expr"
+	"fmt"
+
 	"git.maximhutz.com/max/lambda/pkg/set"
 )
 
 // Expression is the interface for all lambda calculus expression types.
 // It embeds the general expr.Expression interface for cross-mode compatibility.
 type Expression interface {
-	expr.Expression
+	fmt.Stringer
 
 	// Substitute replaces all free occurrences of the target variable with the
 	// replacement expression. Alpha-renaming is performed automatically to
diff --git a/pkg/lambda/marshaler.go b/pkg/lambda/marshaler.go
new file mode 100644
index 0000000..6ee2c79
--- /dev/null
+++ b/pkg/lambda/marshaler.go
@@ -0,0 +1,19 @@
+package lambda
+
+import (
+	"fmt"
+
+	"git.maximhutz.com/max/lambda/pkg/codec"
+)
+
+type Marshaler struct{}
+
+func (m Marshaler) Decode(string) (Expression, error) {
+	return nil, fmt.Errorf("unimplemented")
+}
+
+func (m Marshaler) Encode(e Expression) (string, error) {
+	return e.String(), nil
+}
+
+var _ codec.Marshaler[Expression] = (*Marshaler)(nil)
diff --git a/pkg/normalorder/runtime.go b/pkg/normalorder/runtime.go
deleted file mode 100644
index 37d38e4..0000000
--- a/pkg/normalorder/runtime.go
+++ /dev/null
@@ -1,46 +0,0 @@
-package normalorder
-
-import (
-	"git.maximhutz.com/max/lambda/pkg/emitter"
-	"git.maximhutz.com/max/lambda/pkg/expr"
-	"git.maximhutz.com/max/lambda/pkg/lambda"
-	"git.maximhutz.com/max/lambda/pkg/runtime"
-)
-
-// NormalOrderReducer implements normal order (leftmost-outermost) reduction
-// for lambda calculus expressions.
-type Runtime struct {
-	emitter.BaseEmitter[runtime.Event]
-	expression lambda.Expression
-}
-
-// NewNormalOrderReducer creates a new normal order reducer.
-func NewRuntime(expression lambda.Expression) *Runtime {
-	return &Runtime{
-		BaseEmitter: *emitter.New[runtime.Event](),
-		expression:  expression,
-	}
-}
-
-// Expression returns the current expression state.
-func (r *Runtime) Expression() expr.Expression {
-	return r.expression
-}
-
-func (r *Runtime) Step() bool {
-	result, done := ReduceOnce(r.expression)
-	r.expression = result
-	return !done
-}
-
-// Reduce performs normal order reduction on a lambda expression.
-// The expression must be a lambda.Expression; other types are returned unchanged.
-func (r *Runtime) Run() {
-	r.Emit(runtime.StartEvent)
-
-	for !r.Step() {
-		r.Emit(runtime.StepEvent)
-	}
-
-	r.Emit(runtime.StopEvent)
-}
diff --git a/pkg/runtime/events.go b/pkg/runtime/events.go
deleted file mode 100644
index 6e884c8..0000000
--- a/pkg/runtime/events.go
+++ /dev/null
@@ -1,13 +0,0 @@
-package runtime
-
-// Event represents lifecycle events during interpretation.
-type Event int
-
-const (
-	// StartEvent is emitted before interpretation begins.
-	StartEvent Event = iota
-	// StepEvent is emitted after each interpretation step.
-	StepEvent
-	// StopEvent is emitted after interpretation completes.
-	StopEvent
-)
diff --git a/pkg/runtime/runtime.go b/pkg/runtime/runtime.go
deleted file mode 100644
index c2aafad..0000000
--- a/pkg/runtime/runtime.go
+++ /dev/null
@@ -1,27 +0,0 @@
-// Package runtime provides the abstract Reducer interface for all expression
-// reduction strategies.
-package runtime
-
-import (
-	"git.maximhutz.com/max/lambda/pkg/emitter"
-	"git.maximhutz.com/max/lambda/pkg/expr"
-)
-
-// Runtime defines the interface for expression reduction strategies.
-// Different evaluation modes (normal order, applicative order, SKI combinators,
-// etc.) implement this interface with their own reduction logic.
-//
-// Runtimes also implement the Emitter interface to allow plugins to observe
-// reduction lifecycle events (Start, Step, Stop).
-type Runtime interface {
-	emitter.Emitter[Event]
-
-	// Run a single step. Returns whether the runtime is complete or not.
-	Step() bool
-
-	// Run until completion.
-	Run()
-
-	// Copy the state of the runtime.
-	Expression() expr.Expression
-}
diff --git a/pkg/saccharine/marshaler.go b/pkg/saccharine/marshaler.go
new file mode 100644
index 0000000..bc5c3dc
--- /dev/null
+++ b/pkg/saccharine/marshaler.go
@@ -0,0 +1,24 @@
+// Package "saccharine" provides a simple language built on top of λ-calculus,
+// to facilitate productive coding using it.
+package saccharine
+
+import (
+	"git.maximhutz.com/max/lambda/pkg/codec"
+)
+
+type Marshaler struct{}
+
+func (m Marshaler) Decode(s string) (Expression, error) {
+	tokens, err := scan(s)
+	if err != nil {
+		return nil, err
+	}
+
+	return parse(tokens)
+}
+
+func (m Marshaler) Encode(e Expression) (string, error) {
+	return stringifyExpression(e), nil
+}
+
+var _ codec.Marshaler[Expression] = (*Marshaler)(nil)
diff --git a/pkg/saccharine/parse.go b/pkg/saccharine/parse.go
index eb045d0..8b73394 100644
--- a/pkg/saccharine/parse.go
+++ b/pkg/saccharine/parse.go
@@ -5,18 +5,17 @@ import (
 	"fmt"
 
 	"git.maximhutz.com/max/lambda/pkg/iterator"
-	"git.maximhutz.com/max/lambda/pkg/saccharine/token"
 	"git.maximhutz.com/max/lambda/pkg/trace"
 )
 
-type TokenIterator = iterator.Iterator[token.Token]
+type TokenIterator = iterator.Iterator[Token]
 
-func parseRawToken(i *TokenIterator, expected token.Type) (*token.Token, error) {
-	return iterator.Do(i, func(i *TokenIterator) (*token.Token, error) {
+func parseRawToken(i *TokenIterator, expected TokenType) (*Token, error) {
+	return iterator.Do(i, func(i *TokenIterator) (*Token, error) {
 		if tok, err := i.Next(); err != nil {
 			return nil, err
 		} else if tok.Type != expected {
-			return nil, fmt.Errorf("expected token %v, got %v'", token.Name(expected), tok.Value)
+			return nil, fmt.Errorf("expected token %v, got %v'", expected.Name(), tok.Value)
 		} else {
 			return &tok, nil
 		}
@@ -25,14 +24,14 @@ func parseRawToken(i *TokenIterator, expected token.Type) (*token.Token, error)
 
 func passSoftBreaks(i *TokenIterator) {
 	for {
-		if _, err := parseRawToken(i, token.SoftBreak); err != nil {
+		if _, err := parseRawToken(i, TokenSoftBreak); err != nil {
 			return
 		}
 	}
 }
 
-func parseToken(i *TokenIterator, expected token.Type, ignoreSoftBreaks bool) (*token.Token, error) {
-	return iterator.Do(i, func(i *TokenIterator) (*token.Token, error) {
+func parseToken(i *TokenIterator, expected TokenType, ignoreSoftBreaks bool) (*Token, error) {
+	return iterator.Do(i, func(i *TokenIterator) (*Token, error) {
 		if ignoreSoftBreaks {
 			passSoftBreaks(i)
 		}
@@ -42,17 +41,17 @@ func parseToken(i *TokenIterator, expected token.Type, ignoreSoftBreaks bool) (*
 }
 
 func parseString(i *TokenIterator) (string, error) {
-	if tok, err := parseToken(i, token.Atom, true); err != nil {
+	if tok, err := parseToken(i, TokenAtom, true); err != nil {
 		return "", trace.Wrap(err, "no variable (col %d)", i.Index())
 	} else {
 		return tok.Value, nil
 	}
 }
 
-func parseBreak(i *TokenIterator) (*token.Token, error) {
-	if tok, softErr := parseRawToken(i, token.SoftBreak); softErr == nil {
+func parseBreak(i *TokenIterator) (*Token, error) {
+	if tok, softErr := parseRawToken(i, TokenSoftBreak); softErr == nil {
 		return tok, nil
-	} else if tok, hardErr := parseRawToken(i, token.HardBreak); hardErr == nil {
+	} else if tok, hardErr := parseRawToken(i, TokenHardBreak); hardErr == nil {
 		return tok, nil
 	} else {
 		return nil, errors.Join(softErr, hardErr)
@@ -76,11 +75,11 @@ func parseList[U any](i *TokenIterator, fn func(*TokenIterator) (U, error), mini
 
 func parseAbstraction(i *TokenIterator) (*Abstraction, error) {
 	return iterator.Do(i, func(i *TokenIterator) (*Abstraction, error) {
-		if _, err := parseToken(i, token.Slash, true); err != nil {
+		if _, err := parseToken(i, TokenSlash, true); err != nil {
 			return nil, trace.Wrap(err, "no function slash (col %d)", i.MustGet().Column)
 		} else if parameters, err := parseList(i, parseString, 0); err != nil {
 			return nil, err
-		} else if _, err = parseToken(i, token.Dot, true); err != nil {
+		} else if _, err = parseToken(i, TokenDot, true); err != nil {
 			return nil, trace.Wrap(err, "no function dot (col %d)", i.MustGet().Column)
 		} else if body, err := parseExpression(i); err != nil {
 			return nil, err
@@ -92,11 +91,11 @@ func parseAbstraction(i *TokenIterator) (*Abstraction, error) {
 
 func parseApplication(i *TokenIterator) (*Application, error) {
 	return iterator.Do(i, func(i *TokenIterator) (*Application, error) {
-		if _, err := parseToken(i, token.OpenParen, true); err != nil {
+		if _, err := parseToken(i, TokenOpenParen, true); err != nil {
 			return nil, trace.Wrap(err, "no openning brackets (col %d)", i.MustGet().Column)
 		} else if expressions, err := parseList(i, parseExpression, 1); err != nil {
 			return nil, err
-		} else if _, err := parseToken(i, token.CloseParen, true); err != nil {
+		} else if _, err := parseToken(i, TokenCloseParen, true); err != nil {
 			return nil, trace.Wrap(err, "no closing brackets (col %d)", i.MustGet().Column)
 		} else {
 			return NewApplication(expressions[0], expressions[1:]), nil
@@ -105,7 +104,7 @@ func parseApplication(i *TokenIterator) (*Application, error) {
 }
 
 func parseAtom(i *TokenIterator) (*Atom, error) {
-	if tok, err := parseToken(i, token.Atom, true); err != nil {
+	if tok, err := parseToken(i, TokenAtom, true); err != nil {
 		return nil, trace.Wrap(err, "no variable (col %d)", i.Index())
 	} else {
 		return NewAtom(tok.Value), nil
@@ -133,7 +132,7 @@ func parseStatements(i *TokenIterator) ([]Statement, error) {
 
 func parseClause(i *TokenIterator, braces bool) (*Clause, error) {
 	if braces {
-		if _, err := parseToken(i, token.OpenBrace, true); err != nil {
+		if _, err := parseToken(i, TokenOpenBrace, true); err != nil {
 			return nil, err
 		}
 	}
@@ -152,7 +151,7 @@ func parseClause(i *TokenIterator, braces bool) (*Clause, error) {
 	}
 
 	if braces {
-		if _, err := parseToken(i, token.CloseBrace, true); err != nil {
+		if _, err := parseToken(i, TokenCloseBrace, true); err != nil {
 			return nil, err
 		}
 	}
@@ -165,13 +164,13 @@ func parseExpression(i *TokenIterator) (Expression, error) {
 		passSoftBreaks(i)
 
 		switch peek := i.MustGet(); peek.Type {
-		case token.OpenParen:
+		case TokenOpenParen:
 			return parseApplication(i)
-		case token.Slash:
+		case TokenSlash:
 			return parseAbstraction(i)
-		case token.Atom:
+		case TokenAtom:
 			return parseAtom(i)
-		case token.OpenBrace:
+		case TokenOpenBrace:
 			return parseClause(i, true)
 		default:
 			return nil, fmt.Errorf("expected expression, got '%v' (col %d)", peek.Value, peek.Column)
@@ -183,7 +182,7 @@ func parseLet(i *TokenIterator) (*LetStatement, error) {
 	return iterator.Do(i, func(i *TokenIterator) (*LetStatement, error) {
 		if parameters, err := parseList(i, parseString, 1); err != nil {
 			return nil, err
-		} else if _, err := parseToken(i, token.Assign, true); err != nil {
+		} else if _, err := parseToken(i, TokenAssign, true); err != nil {
 			return nil, err
 		} else if body, err := parseExpression(i); err != nil {
 			return nil, err
@@ -212,7 +211,7 @@ func parseStatement(i *TokenIterator) (Statement, error) {
 }
 
 // Given a list of tokens, attempt to parse it into an syntax tree.
-func parse(tokens []token.Token) (Expression, error) {
+func parse(tokens []Token) (Expression, error) {
 	i := iterator.Of(tokens)
 
 	exp, err := parseClause(i, false)
diff --git a/pkg/saccharine/saccharine.go b/pkg/saccharine/saccharine.go
deleted file mode 100644
index 57b1f17..0000000
--- a/pkg/saccharine/saccharine.go
+++ /dev/null
@@ -1,22 +0,0 @@
-// Package "saccharine" provides a simple language built on top of λ-calculus,
-// to facilitate productive coding using it.
-package saccharine
-
-import (
-	"git.maximhutz.com/max/lambda/pkg/saccharine/token"
-)
-
-// Convert a piece of valid saccharine code into an expression.
-func Parse(code string) (Expression, error) {
-	tokens, err := token.Parse(code)
-	if err != nil {
-		return nil, err
-	}
-
-	return parse(tokens)
-}
-
-// Convert a parsed saccharine expression back into source code.
-func Stringify(expression Expression) string {
-	return stringifyExpression(expression)
-}
diff --git a/pkg/saccharine/token/parse.go b/pkg/saccharine/scan.go
similarity index 70%
rename from pkg/saccharine/token/parse.go
rename to pkg/saccharine/scan.go
index b7d5033..08e3459 100644
--- a/pkg/saccharine/token/parse.go
+++ b/pkg/saccharine/scan.go
@@ -1,4 +1,4 @@
-package token
+package saccharine
 
 import (
 	"errors"
@@ -14,7 +14,7 @@ func isVariable(r rune) bool {
 	return unicode.IsLetter(r) || unicode.IsNumber(r)
 }
 
-func parseRune(i *iterator.Iterator[rune], expected func(rune) bool) (rune, error) {
+func scanRune(i *iterator.Iterator[rune], expected func(rune) bool) (rune, error) {
 	i2 := i.Copy()
 
 	if r, err := i2.Next(); err != nil {
@@ -27,7 +27,7 @@ func parseRune(i *iterator.Iterator[rune], expected func(rune) bool) (rune, erro
 	}
 }
 
-func parseCharacter(i *iterator.Iterator[rune], expected rune) (rune, error) {
+func scanCharacter(i *iterator.Iterator[rune], expected rune) (rune, error) {
 	i2 := i.Copy()
 
 	if r, err := i2.Next(); err != nil {
@@ -42,7 +42,7 @@ func parseCharacter(i *iterator.Iterator[rune], expected rune) (rune, error) {
 
 // Pulls the next token from an iterator over runes. If it cannot, it will
 // return nil. If an error occurs, it will return that.
-func getToken(i *iterator.Iterator[rune]) (*Token, error) {
+func scanToken(i *iterator.Iterator[rune]) (*Token, error) {
 	index := i.Index()
 
 	if i.Done() {
@@ -56,27 +56,27 @@ func getToken(i *iterator.Iterator[rune]) (*Token, error) {
 
 	switch {
 	case letter == '(':
-		return NewOpenParen(index), nil
+		return NewTokenOpenParen(index), nil
 	case letter == ')':
-		return NewCloseParen(index), nil
+		return NewTokenCloseParen(index), nil
 	case letter == '.':
-		return NewDot(index), nil
+		return NewTokenDot(index), nil
 	case letter == '\\':
-		return NewSlash(index), nil
+		return NewTokenSlash(index), nil
 	case letter == '\n':
-		return NewSoftBreak(index), nil
+		return NewTokenSoftBreak(index), nil
 	case letter == '{':
-		return NewOpenBrace(index), nil
+		return NewTokenOpenBrace(index), nil
 	case letter == '}':
-		return NewCloseBrace(index), nil
+		return NewTokenCloseBrace(index), nil
 	case letter == ':':
-		if _, err := parseCharacter(i, '='); err != nil {
+		if _, err := scanCharacter(i, '='); err != nil {
 			return nil, err
 		} else {
-			return NewAssign(index), nil
+			return NewTokenAssign(index), nil
 		}
 	case letter == ';':
-		return NewHardBreak(index), nil
+		return NewTokenHardBreak(index), nil
 	case letter == '#':
 		// Skip everything until the next newline or EOF.
 		for !i.Done() {
@@ -98,27 +98,27 @@ func getToken(i *iterator.Iterator[rune]) (*Token, error) {
 		atom := []rune{letter}
 
 		for {
-			if r, err := parseRune(i, isVariable); err != nil {
+			if r, err := scanRune(i, isVariable); err != nil {
 				break
 			} else {
 				atom = append(atom, r)
 			}
 		}
 
-		return NewAtom(string(atom), index), nil
+		return NewTokenAtom(string(atom), index), nil
 	}
 
 	return nil, fmt.Errorf("unknown character '%v'", string(letter))
 }
 
-// Parse a string into tokens.
-func Parse(input string) ([]Token, error) {
+// scan a string into tokens.
+func scan(input string) ([]Token, error) {
 	i := iterator.Of([]rune(input))
 	tokens := []Token{}
 	errorList := []error{}
 
 	for !i.Done() {
-		token, err := getToken(i)
+		token, err := scanToken(i)
 		if err != nil {
 			errorList = append(errorList, err)
 		} else if token != nil {
diff --git a/pkg/saccharine/token.go b/pkg/saccharine/token.go
new file mode 100644
index 0000000..ee61dbf
--- /dev/null
+++ b/pkg/saccharine/token.go
@@ -0,0 +1,91 @@
+package saccharine
+
+import "fmt"
+
+// All tokens in the pseudo-lambda language.
+type TokenType int
+
+const (
+	TokenOpenParen  TokenType = iota // Denotes the '(' token.
+	TokenCloseParen                  // Denotes the ')' token.
+	TokenOpenBrace                   // Denotes the '{' token.
+	TokenCloseBrace                  // Denotes the '}' token.
+	TokenHardBreak                   // Denotes the ';' token.
+	TokenAssign                      // Denotes the ':=' token.
+	TokenAtom                        // Denotes an alpha-numeric variable.
+	TokenSlash                       // Denotes the '/' token.
+	TokenDot                         // Denotes the '.' token.
+	TokenSoftBreak                   // Denotes a new-line.
+)
+
+// A representation of a token in source code.
+type Token struct {
+	Column int       // Where the token begins in the source text.
+	Type   TokenType // What type the token is.
+	Value  string    // The value of the token.
+}
+
+func NewTokenOpenParen(column int) *Token {
+	return &Token{Type: TokenOpenParen, Column: column, Value: "("}
+}
+
+func NewTokenCloseParen(column int) *Token {
+	return &Token{Type: TokenCloseParen, Column: column, Value: ")"}
+}
+
+func NewTokenOpenBrace(column int) *Token {
+	return &Token{Type: TokenOpenBrace, Column: column, Value: "{"}
+}
+
+func NewTokenCloseBrace(column int) *Token {
+	return &Token{Type: TokenCloseBrace, Column: column, Value: "}"}
+}
+
+func NewTokenDot(column int) *Token {
+	return &Token{Type: TokenDot, Column: column, Value: "."}
+}
+
+func NewTokenHardBreak(column int) *Token {
+	return &Token{Type: TokenHardBreak, Column: column, Value: ";"}
+}
+
+func NewTokenAssign(column int) *Token {
+	return &Token{Type: TokenAssign, Column: column, Value: ":="}
+}
+
+func NewTokenSlash(column int) *Token {
+	return &Token{Type: TokenSlash, Column: column, Value: "\\"}
+}
+
+func NewTokenAtom(name string, column int) *Token {
+	return &Token{Type: TokenAtom, Column: column, Value: name}
+}
+
+func NewTokenSoftBreak(column int) *Token {
+	return &Token{Type: TokenSoftBreak, Column: column, Value: "\\n"}
+}
+
+func (t TokenType) Name() string {
+	switch t {
+	case TokenOpenParen:
+		return "("
+	case TokenCloseParen:
+		return ")"
+	case TokenSlash:
+		return "\\"
+	case TokenDot:
+		return "."
+	case TokenAtom:
+		return "ATOM"
+	case TokenSoftBreak:
+		return "\\n"
+	case TokenHardBreak:
+		return ";"
+	default:
+		panic(fmt.Errorf("unknown token type %v", t))
+	}
+}
+
+func (t Token) Name() string {
+	return t.Type.Name()
+}
diff --git a/pkg/saccharine/token/token.go b/pkg/saccharine/token/token.go
deleted file mode 100644
index 57771a5..0000000
--- a/pkg/saccharine/token/token.go
+++ /dev/null
@@ -1,91 +0,0 @@
-package token
-
-import "fmt"
-
-// All tokens in the pseudo-lambda language.
-type Type int
-
-const (
-	OpenParen  Type = iota // Denotes the '(' token.
-	CloseParen             // Denotes the ')' token.
-	OpenBrace              // Denotes the '{' token.
-	CloseBrace             // Denotes the '}' token.
-	HardBreak              // Denotes the ';' token.
-	Assign                 // Denotes the ':=' token.
-	Atom                   // Denotes an alpha-numeric variable.
-	Slash                  // Denotes the '/' token.
-	Dot                    // Denotes the '.' token.
-	SoftBreak              // Denotes a new-line.
-)
-
-// A representation of a token in source code.
-type Token struct {
-	Column int    // Where the token begins in the source text.
-	Type   Type   // What type the token is.
-	Value  string // The value of the token.
-}
-
-func NewOpenParen(column int) *Token {
-	return &Token{Type: OpenParen, Column: column, Value: "("}
-}
-
-func NewCloseParen(column int) *Token {
-	return &Token{Type: CloseParen, Column: column, Value: ")"}
-}
-
-func NewOpenBrace(column int) *Token {
-	return &Token{Type: OpenBrace, Column: column, Value: "{"}
-}
-
-func NewCloseBrace(column int) *Token {
-	return &Token{Type: CloseBrace, Column: column, Value: "}"}
-}
-
-func NewDot(column int) *Token {
-	return &Token{Type: Dot, Column: column, Value: "."}
-}
-
-func NewHardBreak(column int) *Token {
-	return &Token{Type: HardBreak, Column: column, Value: ";"}
-}
-
-func NewAssign(column int) *Token {
-	return &Token{Type: Assign, Column: column, Value: ":="}
-}
-
-func NewSlash(column int) *Token {
-	return &Token{Type: Slash, Column: column, Value: "\\"}
-}
-
-func NewAtom(name string, column int) *Token {
-	return &Token{Type: Atom, Column: column, Value: name}
-}
-
-func NewSoftBreak(column int) *Token {
-	return &Token{Type: SoftBreak, Column: column, Value: "\\n"}
-}
-
-func Name(typ Type) string {
-	switch typ {
-	case OpenParen:
-		return "("
-	case CloseParen:
-		return ")"
-	case Slash:
-		return "\\"
-	case Dot:
-		return "."
-	case Atom:
-		return "ATOM"
-	case SoftBreak:
-		return "\\n"
-	case HardBreak:
-		return ";"
-	default:
-		panic(fmt.Errorf("unknown token type %v", typ))
-	}
-}
-
-func (t Token) Name() string {
-	return Name(t.Type)
-}