Amolith created
.github/workflows/update.yml | 3
Taskfile.yaml | 5
cmd/neuralwatt/main.go | 250 ++++++++++++++++++++++++
internal/providers/configs/neuralwatt.json | 143 +++++++++++++
internal/providers/providers.go | 8
pkg/catwalk/provider.go | 2
6 files changed, 411 insertions(+)
@@ -25,6 +25,9 @@ jobs:
- run: go run ./cmd/avian/main.go
continue-on-error: true
+ - run: go run ./cmd/neuralwatt/main.go
+ continue-on-error: true
+
- run: go run ./cmd/chutes/main.go
continue-on-error: true
@@ -110,6 +110,11 @@ tasks:
cmds:
- go run cmd/nebius/main.go
+ gen:neuralwatt:
+ desc: Generate Neuralwatt provider configurations
+ cmds:
+ - go run cmd/neuralwatt/main.go
+
gen:openrouter:
desc: Generate openrouter provider configurations
cmds:
@@ -0,0 +1,250 @@
+// Package main provides a command-line tool to fetch models from Neuralwatt
+// and generate a configuration file for the provider.
+package main
+
+import (
+ "context"
+ "encoding/json"
+ "fmt"
+ "io"
+ "log"
+ "math"
+ "net/http"
+ "os"
+ "slices"
+ "strings"
+ "time"
+
+ "charm.land/catwalk/pkg/catwalk"
+)
+
+type NeuralwattModel struct {
+ ID string `json:"id"`
+ MaxModelLen int64 `json:"max_model_len"`
+}
+
+type ModelsResponse struct {
+ Data []NeuralwattModel `json:"data"`
+}
+
+// ModelMeta contains the hardcoded metadata for a Neuralwatt model.
+// The API only returns id and max_model_len, so pricing and capabilities
+// are sourced from the pricing page at https://portal.neuralwatt.com/pricing.
+type ModelMeta struct {
+ Tools bool
+ Reasoning bool
+ Vision bool
+ CostPer1MIn float64
+ CostPer1MOut float64
+}
+
+var modelMetadata = map[string]ModelMeta{
+ "mistralai/Devstral-Small-2-24B-Instruct-2512": {
+ Tools: true,
+ Reasoning: false,
+ Vision: true,
+ CostPer1MIn: 0.1,
+ CostPer1MOut: 0.3,
+ },
+ "zai-org/GLM-5.1-FP8": {
+ Tools: true,
+ Reasoning: true,
+ Vision: false,
+ CostPer1MIn: 1.1,
+ CostPer1MOut: 3.6,
+ },
+ "glm-5.1-fast": {
+ Tools: true,
+ Reasoning: false,
+ Vision: false,
+ CostPer1MIn: 1.1,
+ CostPer1MOut: 3.6,
+ },
+ "openai/gpt-oss-20b": {
+ Tools: true,
+ Reasoning: false,
+ Vision: false,
+ CostPer1MIn: 0.0,
+ CostPer1MOut: 0.2,
+ },
+ "moonshotai/Kimi-K2.5": {
+ Tools: true,
+ Reasoning: false,
+ Vision: true,
+ CostPer1MIn: 0.5,
+ CostPer1MOut: 2.6,
+ },
+ "kimi-k2.5-fast": {
+ Tools: true,
+ Reasoning: false,
+ Vision: true,
+ CostPer1MIn: 0.5,
+ CostPer1MOut: 2.6,
+ },
+ "MiniMaxAI/MiniMax-M2.5": {
+ Tools: true,
+ Reasoning: true,
+ Vision: false,
+ CostPer1MIn: 0.3,
+ CostPer1MOut: 1.4,
+ },
+ "Qwen/Qwen3.5-35B-A3B": {
+ Tools: true,
+ Reasoning: true,
+ Vision: false,
+ CostPer1MIn: 0.3,
+ CostPer1MOut: 1.1,
+ },
+ "Qwen/Qwen3.5-397B-A17B-FP8": {
+ Tools: true,
+ Reasoning: true,
+ Vision: false,
+ CostPer1MIn: 0.7,
+ CostPer1MOut: 4.1,
+ },
+ "qwen3.5-397b-fast": {
+ Tools: true,
+ Reasoning: false,
+ Vision: false,
+ CostPer1MIn: 0.7,
+ CostPer1MOut: 4.1,
+ },
+}
+
+// modelNames provides display names for Neuralwatt-owned models that lack an
+// org prefix and use lowercase IDs.
+var modelNames = map[string]string{
+ "glm-5.1-fast": "GLM 5.1 Fast",
+ "kimi-k2.5-fast": "Kimi K2.5 Fast",
+ "qwen3.5-397b-fast": "Qwen3.5 397B Fast",
+}
+
+func roundCost(v float64) float64 {
+ return math.Round(v*1e5) / 1e5
+}
+
+// modelDisplayName converts a model ID to a human-readable display name. For
+// models with an org prefix (e.g. "zai-org/GLM-5-FP8"), the prefix is stripped.
+// Neuralwatt-owned models without a prefix are looked up in modelNames for
+// proper casing.
+func modelDisplayName(id string) string {
+ if name, ok := modelNames[id]; ok {
+ return name
+ }
+
+ name := id
+ if idx := strings.Index(name, "/"); idx != -1 {
+ name = name[idx+1:]
+ }
+ name = strings.ReplaceAll(name, "-", " ")
+ return name
+}
+
+func fetchNeuralwattModels(apiEndpoint string) (*ModelsResponse, error) {
+ client := &http.Client{Timeout: 30 * time.Second}
+ req, _ := http.NewRequestWithContext(context.Background(), "GET", apiEndpoint+"/models", nil)
+ req.Header.Set("User-Agent", "Crush-Client/1.0")
+
+ resp, err := client.Do(req)
+ if err != nil {
+ return nil, fmt.Errorf("fetching models: %w", err)
+ }
+ defer func() { _ = resp.Body.Close() }()
+
+ body, err := io.ReadAll(resp.Body)
+ if err != nil {
+ return nil, fmt.Errorf("reading models response: %w", err)
+ }
+
+ if resp.StatusCode != 200 {
+ return nil, fmt.Errorf("status %d: %s", resp.StatusCode, body)
+ }
+
+ _ = os.MkdirAll("tmp", 0o700)
+ _ = os.WriteFile("tmp/neuralwatt-response.json", body, 0o600)
+
+ var mr ModelsResponse
+ if err := json.Unmarshal(body, &mr); err != nil {
+ return nil, fmt.Errorf("decoding models response: %w", err)
+ }
+
+ return &mr, nil
+}
+
+func main() {
+ neuralwattProvider := catwalk.Provider{
+ Name: "Neuralwatt",
+ ID: "neuralwatt",
+ APIKey: "$NEURALWATT_API_KEY",
+ APIEndpoint: "https://api.neuralwatt.com/v1",
+ Type: catwalk.TypeOpenAICompat,
+ DefaultLargeModelID: "zai-org/GLM-5.1-FP8",
+ DefaultSmallModelID: "mistralai/Devstral-Small-2-24B-Instruct-2512",
+ }
+
+ modelsResp, err := fetchNeuralwattModels(neuralwattProvider.APIEndpoint)
+ if err != nil {
+ log.Fatal("Error fetching Neuralwatt models:", err)
+ }
+
+ for _, model := range modelsResp.Data {
+ // Skip models with small context windows
+ if model.MaxModelLen < 20000 {
+ fmt.Printf("Skipping model %s: context %d < 20000\n",
+ model.ID, model.MaxModelLen)
+ continue
+ }
+
+ meta, ok := modelMetadata[model.ID]
+ if !ok {
+ fmt.Printf("Skipping unknown model %s (no metadata)\n", model.ID)
+ continue
+ }
+
+ // Only include models that support tools
+ if !meta.Tools {
+ continue
+ }
+
+ var reasoningLevels []string
+ var defaultReasoning string
+ if meta.Reasoning {
+ reasoningLevels = []string{"low", "medium", "high"}
+ defaultReasoning = "medium"
+ }
+
+ m := catwalk.Model{
+ ID: model.ID,
+ Name: modelDisplayName(model.ID),
+ CostPer1MIn: roundCost(meta.CostPer1MIn),
+ CostPer1MOut: roundCost(meta.CostPer1MOut),
+ CostPer1MInCached: 0, // Not available
+ CostPer1MOutCached: 0, // Not available
+ ContextWindow: model.MaxModelLen,
+ DefaultMaxTokens: model.MaxModelLen / 10,
+ CanReason: meta.Reasoning,
+ DefaultReasoningEffort: defaultReasoning,
+ ReasoningLevels: reasoningLevels,
+ SupportsImages: meta.Vision,
+ }
+
+ neuralwattProvider.Models = append(neuralwattProvider.Models, m)
+ fmt.Printf("Added model %s with context window %d\n", model.ID, model.MaxModelLen)
+ }
+
+ slices.SortFunc(neuralwattProvider.Models, func(a catwalk.Model, b catwalk.Model) int {
+ return strings.Compare(a.Name, b.Name)
+ })
+
+ data, err := json.MarshalIndent(neuralwattProvider, "", " ")
+ if err != nil {
+ log.Fatal("Error marshaling Neuralwatt provider:", err)
+ }
+ data = append(data, '\n')
+
+ if err := os.WriteFile("internal/providers/configs/neuralwatt.json", data, 0o600); err != nil {
+ log.Fatal("Error writing Neuralwatt provider config:", err)
+ }
+
+ fmt.Printf("Generated neuralwatt.json with %d models\n", len(neuralwattProvider.Models))
+}
@@ -0,0 +1,143 @@
+{
+ "name": "Neuralwatt",
+ "id": "neuralwatt",
+ "api_key": "$NEURALWATT_API_KEY",
+ "api_endpoint": "https://api.neuralwatt.com/v1",
+ "type": "openai-compat",
+ "default_large_model_id": "zai-org/GLM-5.1-FP8",
+ "default_small_model_id": "mistralai/Devstral-Small-2-24B-Instruct-2512",
+ "models": [
+ {
+ "id": "mistralai/Devstral-Small-2-24B-Instruct-2512",
+ "name": "Devstral Small 2 24B Instruct 2512",
+ "cost_per_1m_in": 0.1,
+ "cost_per_1m_out": 0.3,
+ "cost_per_1m_in_cached": 0,
+ "cost_per_1m_out_cached": 0,
+ "context_window": 262144,
+ "default_max_tokens": 26214,
+ "can_reason": false,
+ "supports_attachments": true
+ },
+ {
+ "id": "zai-org/GLM-5.1-FP8",
+ "name": "GLM 5.1 FP8",
+ "cost_per_1m_in": 1.1,
+ "cost_per_1m_out": 3.6,
+ "cost_per_1m_in_cached": 0,
+ "cost_per_1m_out_cached": 0,
+ "context_window": 202752,
+ "default_max_tokens": 20275,
+ "can_reason": true,
+ "reasoning_levels": [
+ "low",
+ "medium",
+ "high"
+ ],
+ "default_reasoning_effort": "medium",
+ "supports_attachments": false
+ },
+ {
+ "id": "glm-5.1-fast",
+ "name": "GLM 5.1 Fast",
+ "cost_per_1m_in": 1.1,
+ "cost_per_1m_out": 3.6,
+ "cost_per_1m_in_cached": 0,
+ "cost_per_1m_out_cached": 0,
+ "context_window": 202752,
+ "default_max_tokens": 20275,
+ "can_reason": false,
+ "supports_attachments": false
+ },
+ {
+ "id": "moonshotai/Kimi-K2.5",
+ "name": "Kimi K2.5",
+ "cost_per_1m_in": 0.5,
+ "cost_per_1m_out": 2.6,
+ "cost_per_1m_in_cached": 0,
+ "cost_per_1m_out_cached": 0,
+ "context_window": 262144,
+ "default_max_tokens": 26214,
+ "can_reason": false,
+ "supports_attachments": true
+ },
+ {
+ "id": "kimi-k2.5-fast",
+ "name": "Kimi K2.5 Fast",
+ "cost_per_1m_in": 0.5,
+ "cost_per_1m_out": 2.6,
+ "cost_per_1m_in_cached": 0,
+ "cost_per_1m_out_cached": 0,
+ "context_window": 262144,
+ "default_max_tokens": 26214,
+ "can_reason": false,
+ "supports_attachments": true
+ },
+ {
+ "id": "MiniMaxAI/MiniMax-M2.5",
+ "name": "MiniMax M2.5",
+ "cost_per_1m_in": 0.3,
+ "cost_per_1m_out": 1.4,
+ "cost_per_1m_in_cached": 0,
+ "cost_per_1m_out_cached": 0,
+ "context_window": 196608,
+ "default_max_tokens": 19660,
+ "can_reason": true,
+ "reasoning_levels": [
+ "low",
+ "medium",
+ "high"
+ ],
+ "default_reasoning_effort": "medium",
+ "supports_attachments": false
+ },
+ {
+ "id": "Qwen/Qwen3.5-35B-A3B",
+ "name": "Qwen3.5 35B A3B",
+ "cost_per_1m_in": 0.3,
+ "cost_per_1m_out": 1.1,
+ "cost_per_1m_in_cached": 0,
+ "cost_per_1m_out_cached": 0,
+ "context_window": 131072,
+ "default_max_tokens": 13107,
+ "can_reason": true,
+ "reasoning_levels": [
+ "low",
+ "medium",
+ "high"
+ ],
+ "default_reasoning_effort": "medium",
+ "supports_attachments": false
+ },
+ {
+ "id": "Qwen/Qwen3.5-397B-A17B-FP8",
+ "name": "Qwen3.5 397B A17B FP8",
+ "cost_per_1m_in": 0.7,
+ "cost_per_1m_out": 4.1,
+ "cost_per_1m_in_cached": 0,
+ "cost_per_1m_out_cached": 0,
+ "context_window": 262144,
+ "default_max_tokens": 26214,
+ "can_reason": true,
+ "reasoning_levels": [
+ "low",
+ "medium",
+ "high"
+ ],
+ "default_reasoning_effort": "medium",
+ "supports_attachments": false
+ },
+ {
+ "id": "qwen3.5-397b-fast",
+ "name": "Qwen3.5 397B Fast",
+ "cost_per_1m_in": 0.7,
+ "cost_per_1m_out": 4.1,
+ "cost_per_1m_in_cached": 0,
+ "cost_per_1m_out_cached": 0,
+ "context_window": 262144,
+ "default_max_tokens": 26214,
+ "can_reason": false,
+ "supports_attachments": false
+ }
+ ]
+}
@@ -96,6 +96,9 @@ var qiniuCloudConfig []byte
//go:embed configs/avian.json
var avianConfig []byte
+//go:embed configs/neuralwatt.json
+var neuralwattConfig []byte
+
// ProviderFunc is a function that returns a Provider.
type ProviderFunc func() catwalk.Provider
@@ -129,6 +132,7 @@ var providerRegistry = []ProviderFunc{
qiniuCloudProvider,
avianProvider,
nebiusProvider,
+ neuralwattProvider,
}
// GetAll returns all registered providers.
@@ -264,3 +268,7 @@ func nebiusProvider() catwalk.Provider {
func avianProvider() catwalk.Provider {
return loadProviderFromConfig(avianConfig)
}
+
+func neuralwattProvider() catwalk.Provider {
+ return loadProviderFromConfig(neuralwattConfig)
+}
@@ -49,6 +49,7 @@ const (
InferenceProviderQiniuCloud InferenceProvider = "qiniucloud"
InferenceProviderAvian InferenceProvider = "avian"
InferenceProviderNebius InferenceProvider = "nebius"
+ InferenceProviderNeuralwatt InferenceProvider = "neuralwatt"
)
// Provider represents an AI provider configuration.
@@ -121,6 +122,7 @@ func KnownProviders() []InferenceProvider {
InferenceProviderQiniuCloud,
InferenceProviderAvian,
InferenceProviderNebius,
+ InferenceProviderNeuralwatt,
}
}