provider.go

v0.2.0

package provider

// StaticModel represents a model available from a provider (no API fetch needed).
type StaticModel struct {
	ID     string
	Name   string
	Family string
}

// Plan describes one billing/access plan for a provider with multiple base URLs.
type Plan struct {
	Key         string // e.g. "api", "coding"
	Name        string // display name, e.g. "API (pay per token)"
	Description string
	BaseURL     string
}

var zaiModels = []StaticModel{
	{ID: "GLM-5", Name: "GLM-5", Family: "GLM-5"},
	{ID: "GLM-4.7", Name: "GLM-4.7", Family: "GLM-4.7"},
	{ID: "GLM-4.6", Name: "GLM-4.6", Family: "GLM-4.6"},
	{ID: "GLM-4.5", Name: "GLM-4.5", Family: "GLM-4.5"},
	{ID: "GLM-4.5-Air", Name: "GLM-4.5 Air", Family: "GLM-4.5"},
	{ID: "GLM-4.5-X", Name: "GLM-4.5 X", Family: "GLM-4.5"},
	{ID: "GLM-4.5-AirX", Name: "GLM-4.5 AirX", Family: "GLM-4.5"},
	{ID: "GLM-4.5-Flash", Name: "GLM-4.5 Flash", Family: "GLM-4.5"},
	{ID: "GLM-4-32B-0414-128K", Name: "GLM-4 32B", Family: "GLM-4"},
}

// StaticModels maps provider keys to their available models.
var StaticModels = map[string][]StaticModel{
	"deepseek": {
		{ID: "deepseek-reasoner", Name: "DeepSeek Reasoner (R1)", Family: "Reasoner"},
		{ID: "deepseek-chat", Name: "DeepSeek Chat (V3)", Family: "Chat"},
	},
	"zai": zaiModels,
	"kimi": {
		{ID: "kimi-k2.5", Name: "Kimi K2.5", Family: "K2.5"},
		{ID: "kimi-k2-turbo-preview", Name: "Kimi K2 Turbo", Family: "K2"},
		{ID: "kimi-k2-thinking", Name: "Kimi K2 Thinking", Family: "K2"},
		{ID: "kimi-k2-0711-preview", Name: "Kimi K2", Family: "K2"},
		{ID: "kimi-latest", Name: "Kimi Latest", Family: "Latest"},
	},
}

// Provider defines an alternative Claude Code API provider.
type Provider struct {
	Name        string
	Description string
	BaseURL     string
	Model       string
	SmallModel  string
	SonnetModel string
	OpusModel   string
	HaikuModel  string
	TimeoutMS   string
	// AuthEnvVar is the env var name used for the API key (some use AUTH_TOKEN, some API_KEY).
	AuthEnvVar string
	// Compat is "anthropic" (default) or "openai". When "openai", a local
	// translation proxy is started to convert Anthropic ↔ OpenAI API format.
	Compat string
	// ContextWindow is the model's context window size in tokens.
	// Claude Code assumes 200k by default; when this is smaller,
	// CLAUDE_AUTOCOMPACT_PCT_OVERRIDE is set so compaction triggers before
	// the real limit is hit.
	ContextWindow int
	// NoAuth indicates the provider does not require an API key (e.g. local Ollama).
	NoAuth bool
	// Plans lists alternative billing/access plans with different base URLs.
	// When non-empty, the TUI prompts the user to choose a plan.
	// BaseURL holds the default plan's URL.
	Plans []Plan
}

var Registry = map[string]Provider{
	"deepseek": {
		Name:          "DeepSeek",
		Description:   "DeepSeek Reasoner / V3 via Anthropic-compatible API",
		BaseURL:       "https://api.deepseek.com/anthropic",
		Model:         "deepseek-reasoner",
		SmallModel:    "deepseek-chat",
		SonnetModel:   "deepseek-reasoner",
		OpusModel:     "deepseek-reasoner",
		HaikuModel:    "deepseek-chat",
		TimeoutMS:     "600000",
		AuthEnvVar:    "ANTHROPIC_AUTH_TOKEN",
		ContextWindow: 128_000,
	},
	"zai": {
		Name:          "z.AI",
		Description:   "GLM-5 / GLM-4.7 via z.AI (API or Coding Plan)",
		BaseURL:       "https://api.z.ai/api/anthropic",
		Model:         "GLM-5",
		SmallModel:    "GLM-4.5-Air",
		SonnetModel:   "GLM-5",
		OpusModel:     "GLM-5",
		HaikuModel:    "GLM-4.5-Air",
		TimeoutMS:     "3000000",
		AuthEnvVar:    "ANTHROPIC_AUTH_TOKEN",
		ContextWindow: 205_000,
		Plans: []Plan{
			{Key: "api", Name: "API (pay per token)", Description: "Pay-per-token access", BaseURL: "https://api.z.ai/api/anthropic"},
			{Key: "coding", Name: "Coding Plan (subscription)", Description: "Subscription-based access", BaseURL: "https://api.z.ai/api/coding/anthropic"},
		},
	},
	"minimax": {
		Name:          "MiniMax",
		Description:   "MiniMax-M2.5 via MiniMax Anthropic gateway",
		BaseURL:       "https://api.minimax.io/anthropic",
		Model:         "MiniMax-M2.5",
		SmallModel:    "MiniMax-M2.5",
		SonnetModel:   "MiniMax-M2.5",
		OpusModel:     "MiniMax-M2.5",
		HaikuModel:    "MiniMax-M2.5",
		TimeoutMS:     "3000000",
		AuthEnvVar:    "ANTHROPIC_AUTH_TOKEN",
		ContextWindow: 205_000,
	},
	"kimi": {
		Name:          "Kimi",
		Description:   "Kimi K2.5 / K2 via Moonshot Anthropic gateway",
		BaseURL:       "https://api.moonshot.cn/anthropic/",
		Model:         "kimi-k2.5",
		SmallModel:    "kimi-k2-turbo-preview",
		SonnetModel:   "kimi-k2.5",
		OpusModel:     "kimi-k2.5",
		HaikuModel:    "kimi-k2-turbo-preview",
		TimeoutMS:     "600000",
		AuthEnvVar:    "ANTHROPIC_AUTH_TOKEN",
		ContextWindow: 256_000,
	},
	"copilot": {
		Name:          "GitHub Copilot",
		Description:   "Claude via GitHub Copilot",
		BaseURL:       "https://api.githubcopilot.com",
		Model:         "claude-sonnet-4.6",
		SmallModel:    "claude-haiku-4.5",
		SonnetModel:   "claude-sonnet-4.6",
		OpusModel:     "claude-opus-4.6",
		HaikuModel:    "claude-haiku-4.5",
		TimeoutMS:     "600000",
		AuthEnvVar:    "ANTHROPIC_AUTH_TOKEN",
		Compat:        "openai",
		ContextWindow: 200_000,
	},
	"nvidia": {
		Name:          "NVIDIA NIM",
		Description:   "Access NVIDIA NIM models via OpenAI-compatible API",
		BaseURL:       "https://integrate.api.nvidia.com/v1",
		Model:         "nvidia/llama-3.1-nemotron-ultra-253b-v1",
		SmallModel:    "nvidia/llama-3.3-nemotron-super-49b-v1.5",
		SonnetModel:   "nvidia/llama-3.1-nemotron-ultra-253b-v1",
		OpusModel:     "nvidia/llama-3.1-nemotron-ultra-253b-v1",
		HaikuModel:    "nvidia/llama-3.3-nemotron-super-49b-v1.5",
		TimeoutMS:     "600000",
		AuthEnvVar:    "ANTHROPIC_AUTH_TOKEN",
		Compat:        "openai",
		ContextWindow: 128_000,
	},
	"openrouter": {
		Name:          "OpenRouter",
		Description:   "Access multiple models via OpenRouter",
		BaseURL:       "https://openrouter.ai/api/v1",
		Model:         "anthropic/claude-sonnet-4.6",
		SmallModel:    "anthropic/claude-haiku-4.5",
		SonnetModel:   "anthropic/claude-sonnet-4.6",
		OpusModel:     "anthropic/claude-opus-4.6",
		HaikuModel:    "anthropic/claude-haiku-4.5",
		TimeoutMS:     "600000",
		AuthEnvVar:    "ANTHROPIC_AUTH_TOKEN",
		Compat:        "openai",
		ContextWindow: 200_000,
	},
	"vkproxy": {
		Name:          "VK LLM Proxy",
		Description:   "Import models from claude-code-config.zip",
		BaseURL:       "https://llm-proxy.vkteam.ru",
		Model:         "deepseek-reasoner",
		SmallModel:    "deepseek-reasoner",
		SonnetModel:   "deepseek-reasoner",
		OpusModel:     "deepseek-reasoner",
		HaikuModel:    "deepseek-reasoner",
		TimeoutMS:     "600000",
		AuthEnvVar:    "ANTHROPIC_AUTH_TOKEN",
		ContextWindow: 128_000,
	},
	"ollama": {
		Name:          "Ollama (local)",
		Description:   "Local Ollama with native Anthropic-compatible API",
		BaseURL:       "http://localhost:11434",
		Model:         "qwen3-coder",
		SmallModel:    "qwen3-coder",
		SonnetModel:   "qwen3-coder",
		OpusModel:     "qwen3-coder",
		HaikuModel:    "qwen3-coder",
		TimeoutMS:     "600000",
		AuthEnvVar:    "ANTHROPIC_AUTH_TOKEN",
		ContextWindow: 32_000,
		NoAuth:        true,
	},
	"ollama-cloud": {
		Name:          "Ollama Cloud",
		Description:   "Cloud models via ollama.com API",
		BaseURL:       "https://ollama.com",
		Model:         "qwen3.5:397b",
		SmallModel:    "devstral-small-2:24b",
		SonnetModel:   "devstral-2:123b",
		OpusModel:     "qwen3.5:397b",
		HaikuModel:    "devstral-small-2:24b",
		TimeoutMS:     "600000",
		AuthEnvVar:    "ANTHROPIC_AUTH_TOKEN",
		ContextWindow: 32_000,
	},
}

// Order defines the display order of providers.
var Order = []string{"deepseek", "zai", "minimax", "kimi", "copilot", "nvidia", "openrouter", "ollama", "ollama-cloud", "vkproxy"}


Source Files

provider.go

1	package provider
2
3	// StaticModel represents a model available from a provider (no API fetch needed).
4	type StaticModel struct {
5	ID string
6	Name string
7	Family string
8	}
9
10	// Plan describes one billing/access plan for a provider with multiple base URLs.
11	type Plan struct {
12	Key string // e.g. "api", "coding"
13	Name string // display name, e.g. "API (pay per token)"
14	Description string
15	BaseURL string
16	}
17
18	var zaiModels = []StaticModel{
19	{ID: "GLM-5", Name: "GLM-5", Family: "GLM-5"},
20	{ID: "GLM-4.7", Name: "GLM-4.7", Family: "GLM-4.7"},
21	{ID: "GLM-4.6", Name: "GLM-4.6", Family: "GLM-4.6"},
22	{ID: "GLM-4.5", Name: "GLM-4.5", Family: "GLM-4.5"},
23	{ID: "GLM-4.5-Air", Name: "GLM-4.5 Air", Family: "GLM-4.5"},
24	{ID: "GLM-4.5-X", Name: "GLM-4.5 X", Family: "GLM-4.5"},
25	{ID: "GLM-4.5-AirX", Name: "GLM-4.5 AirX", Family: "GLM-4.5"},
26	{ID: "GLM-4.5-Flash", Name: "GLM-4.5 Flash", Family: "GLM-4.5"},
27	{ID: "GLM-4-32B-0414-128K", Name: "GLM-4 32B", Family: "GLM-4"},
28	}
29
30	// StaticModels maps provider keys to their available models.
31	var StaticModels = map[string][]StaticModel{
32	"deepseek": {
33	{ID: "deepseek-reasoner", Name: "DeepSeek Reasoner (R1)", Family: "Reasoner"},
34	{ID: "deepseek-chat", Name: "DeepSeek Chat (V3)", Family: "Chat"},
35	},
36	"zai": zaiModels,
37	"kimi": {
38	{ID: "kimi-k2.5", Name: "Kimi K2.5", Family: "K2.5"},
39	{ID: "kimi-k2-turbo-preview", Name: "Kimi K2 Turbo", Family: "K2"},
40	{ID: "kimi-k2-thinking", Name: "Kimi K2 Thinking", Family: "K2"},
41	{ID: "kimi-k2-0711-preview", Name: "Kimi K2", Family: "K2"},
42	{ID: "kimi-latest", Name: "Kimi Latest", Family: "Latest"},
43	},
44	}
45
46	// Provider defines an alternative Claude Code API provider.
47	type Provider struct {
48	Name string
49	Description string
50	BaseURL string
51	Model string
52	SmallModel string
53	SonnetModel string
54	OpusModel string
55	HaikuModel string
56	TimeoutMS string
57	// AuthEnvVar is the env var name used for the API key (some use AUTH_TOKEN, some API_KEY).
58	AuthEnvVar string
59	// Compat is "anthropic" (default) or "openai". When "openai", a local
60	// translation proxy is started to convert Anthropic ↔ OpenAI API format.
61	Compat string
62	// ContextWindow is the model's context window size in tokens.
63	// Claude Code assumes 200k by default; when this is smaller,
64	// CLAUDE_AUTOCOMPACT_PCT_OVERRIDE is set so compaction triggers before
65	// the real limit is hit.
66	ContextWindow int
67	// NoAuth indicates the provider does not require an API key (e.g. local Ollama).
68	NoAuth bool
69	// Plans lists alternative billing/access plans with different base URLs.
70	// When non-empty, the TUI prompts the user to choose a plan.
71	// BaseURL holds the default plan's URL.
72	Plans []Plan
73	}
74
75	var Registry = map[string]Provider{
76	"deepseek": {
77	Name: "DeepSeek",
78	Description: "DeepSeek Reasoner / V3 via Anthropic-compatible API",
79	BaseURL: "https://api.deepseek.com/anthropic",
80	Model: "deepseek-reasoner",
81	SmallModel: "deepseek-chat",
82	SonnetModel: "deepseek-reasoner",
83	OpusModel: "deepseek-reasoner",
84	HaikuModel: "deepseek-chat",
85	TimeoutMS: "600000",
86	AuthEnvVar: "ANTHROPIC_AUTH_TOKEN",
87	ContextWindow: 128_000,
88	},
89	"zai": {
90	Name: "z.AI",
91	Description: "GLM-5 / GLM-4.7 via z.AI (API or Coding Plan)",
92	BaseURL: "https://api.z.ai/api/anthropic",
93	Model: "GLM-5",
94	SmallModel: "GLM-4.5-Air",
95	SonnetModel: "GLM-5",
96	OpusModel: "GLM-5",
97	HaikuModel: "GLM-4.5-Air",
98	TimeoutMS: "3000000",
99	AuthEnvVar: "ANTHROPIC_AUTH_TOKEN",
100	ContextWindow: 205_000,
101	Plans: []Plan{
102	{Key: "api", Name: "API (pay per token)", Description: "Pay-per-token access", BaseURL: "https://api.z.ai/api/anthropic"},
103	{Key: "coding", Name: "Coding Plan (subscription)", Description: "Subscription-based access", BaseURL: "https://api.z.ai/api/coding/anthropic"},
104	},
105	},
106	"minimax": {
107	Name: "MiniMax",
108	Description: "MiniMax-M2.5 via MiniMax Anthropic gateway",
109	BaseURL: "https://api.minimax.io/anthropic",
110	Model: "MiniMax-M2.5",
111	SmallModel: "MiniMax-M2.5",
112	SonnetModel: "MiniMax-M2.5",
113	OpusModel: "MiniMax-M2.5",
114	HaikuModel: "MiniMax-M2.5",
115	TimeoutMS: "3000000",
116	AuthEnvVar: "ANTHROPIC_AUTH_TOKEN",
117	ContextWindow: 205_000,
118	},
119	"kimi": {
120	Name: "Kimi",
121	Description: "Kimi K2.5 / K2 via Moonshot Anthropic gateway",
122	BaseURL: "https://api.moonshot.cn/anthropic/",
123	Model: "kimi-k2.5",
124	SmallModel: "kimi-k2-turbo-preview",
125	SonnetModel: "kimi-k2.5",
126	OpusModel: "kimi-k2.5",
127	HaikuModel: "kimi-k2-turbo-preview",
128	TimeoutMS: "600000",
129	AuthEnvVar: "ANTHROPIC_AUTH_TOKEN",
130	ContextWindow: 256_000,
131	},
132	"copilot": {
133	Name: "GitHub Copilot",
134	Description: "Claude via GitHub Copilot",
135	BaseURL: "https://api.githubcopilot.com",
136	Model: "claude-sonnet-4.6",
137	SmallModel: "claude-haiku-4.5",
138	SonnetModel: "claude-sonnet-4.6",
139	OpusModel: "claude-opus-4.6",
140	HaikuModel: "claude-haiku-4.5",
141	TimeoutMS: "600000",
142	AuthEnvVar: "ANTHROPIC_AUTH_TOKEN",
143	Compat: "openai",
144	ContextWindow: 200_000,
145	},
146	"nvidia": {
147	Name: "NVIDIA NIM",
148	Description: "Access NVIDIA NIM models via OpenAI-compatible API",
149	BaseURL: "https://integrate.api.nvidia.com/v1",
150	Model: "nvidia/llama-3.1-nemotron-ultra-253b-v1",
151	SmallModel: "nvidia/llama-3.3-nemotron-super-49b-v1.5",
152	SonnetModel: "nvidia/llama-3.1-nemotron-ultra-253b-v1",
153	OpusModel: "nvidia/llama-3.1-nemotron-ultra-253b-v1",
154	HaikuModel: "nvidia/llama-3.3-nemotron-super-49b-v1.5",
155	TimeoutMS: "600000",
156	AuthEnvVar: "ANTHROPIC_AUTH_TOKEN",
157	Compat: "openai",
158	ContextWindow: 128_000,
159	},
160	"openrouter": {
161	Name: "OpenRouter",
162	Description: "Access multiple models via OpenRouter",
163	BaseURL: "https://openrouter.ai/api/v1",
164	Model: "anthropic/claude-sonnet-4.6",
165	SmallModel: "anthropic/claude-haiku-4.5",
166	SonnetModel: "anthropic/claude-sonnet-4.6",
167	OpusModel: "anthropic/claude-opus-4.6",
168	HaikuModel: "anthropic/claude-haiku-4.5",
169	TimeoutMS: "600000",
170	AuthEnvVar: "ANTHROPIC_AUTH_TOKEN",
171	Compat: "openai",
172	ContextWindow: 200_000,
173	},
174	"vkproxy": {
175	Name: "VK LLM Proxy",
176	Description: "Import models from claude-code-config.zip",
177	BaseURL: "https://llm-proxy.vkteam.ru",
178	Model: "deepseek-reasoner",
179	SmallModel: "deepseek-reasoner",
180	SonnetModel: "deepseek-reasoner",
181	OpusModel: "deepseek-reasoner",
182	HaikuModel: "deepseek-reasoner",
183	TimeoutMS: "600000",
184	AuthEnvVar: "ANTHROPIC_AUTH_TOKEN",
185	ContextWindow: 128_000,
186	},
187	"ollama": {
188	Name: "Ollama (local)",
189	Description: "Local Ollama with native Anthropic-compatible API",
190	BaseURL: "http://localhost:11434",
191	Model: "qwen3-coder",
192	SmallModel: "qwen3-coder",
193	SonnetModel: "qwen3-coder",
194	OpusModel: "qwen3-coder",
195	HaikuModel: "qwen3-coder",
196	TimeoutMS: "600000",
197	AuthEnvVar: "ANTHROPIC_AUTH_TOKEN",
198	ContextWindow: 32_000,
199	NoAuth: true,
200	},
201	"ollama-cloud": {
202	Name: "Ollama Cloud",
203	Description: "Cloud models via ollama.com API",
204	BaseURL: "https://ollama.com",
205	Model: "qwen3.5:397b",
206	SmallModel: "devstral-small-2:24b",
207	SonnetModel: "devstral-2:123b",
208	OpusModel: "qwen3.5:397b",
209	HaikuModel: "devstral-small-2:24b",
210	TimeoutMS: "600000",
211	AuthEnvVar: "ANTHROPIC_AUTH_TOKEN",
212	ContextWindow: 32_000,
213	},
214	}
215
216	// Order defines the display order of providers.
217	var Order = []string{"deepseek", "zai", "minimax", "kimi", "copilot", "nvidia", "openrouter", "ollama", "ollama-cloud", "vkproxy"}
218