foundry-core/internal/codex/canvas.go

package codex

import (
	"context"
	"crypto/sha256"
	"encoding/hex"
	"encoding/json"
	"errors"
	"io"
	"net"
	"net/http"
	"os"
	"sort"
	"strings"
	"time"
)

const CanvasInputLimit = 1 << 20

type CanvasModel struct {
	ID               string         `json:"id"`
	Label            string         `json:"label"`
	Provider         string         `json:"provider"`
	InputModalities  []string       `json:"input_modalities"`
	OutputModalities []string       `json:"output_modalities"`
	Available        bool           `json:"available"`
	MaxExecutionMS   int            `json:"max_execution_ms"`
	MaxInputBytes    int            `json:"max_input_bytes"`
	ParametersSchema map[string]any `json:"parameters_schema"`
	Backend          string         `json:"-"`
}

// Only curated catalog fields leave the worker. Raw CLI cache metadata,
// local paths, authentication files and backend configuration stay private.
func workerCanvasModels() ([]CanvasModel, error) {
	raw, err := os.ReadFile("/home/node/.codex/models_cache.json")
	if err != nil {
		return nil, err
	}
	var cache struct {
		Models []struct {
			Slug       string   `json:"slug"`
			Label      string   `json:"display_name"`
			Visibility string   `json:"visibility"`
			Inputs     []string `json:"input_modalities"`
		} `json:"models"`
	}
	if err = json.Unmarshal(raw, &cache); err != nil {
		return nil, err
	}
	result := []CanvasModel{}
	for _, entry := range cache.Models {
		if entry.Visibility != "list" || entry.Slug == "" || strings.HasPrefix(entry.Slug, "-") || len(entry.Slug) > 128 {
			continue
		}
		sum := sha256.Sum256([]byte("codex-cli/" + entry.Slug))
		inputs := []string{}
		for _, value := range entry.Inputs {
			if value == "text" || value == "image" {
				inputs = append(inputs, value)
			}
		}
		if len(inputs) == 0 {
			continue
		}
		label := entry.Label
		if label == "" {
			label = entry.Slug
		}
		if len(label) > 128 {
			label = entry.Slug
		}
		result = append(result, CanvasModel{ID: "codex-" + hex.EncodeToString(sum[:16]), Label: label, Provider: "codex-cli", InputModalities: inputs, OutputModalities: []string{"text"}, Available: true, MaxExecutionMS: 90000, MaxInputBytes: CanvasInputLimit, ParametersSchema: map[string]any{"type": "object", "properties": map[string]any{}, "additionalProperties": false}, Backend: entry.Slug})
	}
	verified := map[string]bool{}
	for _, id := range strings.Split(os.Getenv("FOUNDRY_CANVAS_IMAGE_MODELS"), ",") {
		id = strings.TrimSpace(id)
		if id != "" {
			verified[id] = true
		}
	}
	textModels := append([]CanvasModel{}, result...)
	for _, model := range textModels {
		if !verified[model.Backend] {
			continue
		}
		sum := sha256.Sum256([]byte("codex-cli/" + model.Backend + "/image"))
		model.ID = "codex-" + hex.EncodeToString(sum[:16])
		model.Label += " · Image"
		model.OutputModalities = []string{"image"}
		model.MaxExecutionMS = 180000
		result = append(result, model)
	}
	sort.Slice(result, func(i, j int) bool { return result[i].ID < result[j].ID })
	return result, nil
}

type CanvasClient struct{ client *http.Client }

func NewCanvasClient(socket string) *CanvasClient {
	if socket == "" {
		return nil
	}
	transport := &http.Transport{DisableKeepAlives: true, DialContext: func(ctx context.Context, _, _ string) (net.Conn, error) {
		return (&net.Dialer{}).DialContext(ctx, "unix", socket)
	}}
	return &CanvasClient{client: &http.Client{Transport: transport, Timeout: 182 * time.Second, CheckRedirect: func(*http.Request, []*http.Request) error { return http.ErrUseLastResponse }}}
}
func (c *CanvasClient) Models(ctx context.Context) ([]CanvasModel, error) {
	if c == nil {
		return nil, errors.New("codex_unavailable")
	}
	request, err := http.NewRequestWithContext(ctx, "GET", "http://codex/canvas/models", nil)
	if err != nil {
		return nil, err
	}
	response, err := c.client.Do(request)
	if err != nil {
		return nil, err
	}
	defer response.Body.Close()
	if response.StatusCode != 200 {
		return nil, errors.New("codex_unavailable")
	}
	raw, err := io.ReadAll(io.LimitReader(response.Body, 256<<10))
	if err != nil {
		return nil, err
	}
	var result struct {
		Models []CanvasModel `json:"models"`
	}
	if err = json.Unmarshal(raw, &result); err != nil {
		return nil, err
	}
	return result.Models, nil
}

type CanvasTask struct {
	OutputModality string         `json:"output_modality,omitempty"`
	Images         []CanvasImage  `json:"images,omitempty"`
	ModelID        string         `json:"model_id"`
	Prompt         string         `json:"prompt"`
	Parameters     map[string]any `json:"parameters"`
}

func (c *CanvasClient) Generate(ctx context.Context, input CanvasTask) (string, error) {
	output, err := c.GenerateOutput(ctx, input)
	if err != nil {
		return "", err
	}
	if output.Kind != "text" {
		return "", errors.New("model_incompatible")
	}
	return output.Text, nil
}