@magpie-community/opencode-factory-auth 0.1.30 → 0.1.32

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (3) hide show
  1. package/README.md +8 -0
  2. package/index.mjs +139 -51
  3. package/package.json +1 -1
package/README.md CHANGED
@@ -73,6 +73,14 @@ Flash, Kimi K3 and GLM-5.2 have `none`, droid's "off": they stop thinking.
73
73
  Reasoning is mandatory on GLM-5.3 and GLM-5.3-Flash (Fireworks turns `none`
74
74
  away), so `low` is their least.
75
75
 
76
+ A model's price is its token multiplier in droid's registry
77
+ (`cost.tokenMultiplier`), which magpie shows beside the model as its rate
78
+ (`0.8×`), so a cheap one is picked without opening droid. A promotion running
79
+ comes off it, with the multiplier before the discount struck through beside
80
+ the price (`rateWas`): Mistral Large 4's launch discount, Gemini 4 Argon's,
81
+ Gemini 3.8/3.7 Flash's and GPT-5.6 Sol's, each while droid's own dates for it
82
+ hold. A promotion that has ended is no discount at all.
83
+
76
84
  Gemini's requests go to Factory's generate route with the model in the
77
85
  body, droid's line opening the system instruction. Gemini 4 Argon is behind
78
86
  a Factory feature flag in droid 0.241.1, so an account it isn't on for may
package/index.mjs CHANGED
@@ -56,62 +56,62 @@ const E4 = ["low", "medium", "high", "xhigh"]
56
56
  // base models aren't offered), GPT-5.6 Sol Fast (unreleased:!0) and
57
57
  // GLM-5.2 Fast (behind a feature flag, on a Baseten model id of its own).
58
58
  const MODELS = [
59
- ["claude-fable-5.1", "Fable 5.1", ANTHROPIC, "anthropic", 867000, 128000, E5, true],
60
- ["claude-fable-5", "Fable 5", ANTHROPIC, "anthropic", 867000, 128000, E5, true],
61
- ["claude-opus-5-5", "Opus 5.5", ANTHROPIC, "anthropic", 872000, 128000, E5, true],
62
- ["claude-opus-5", "Opus 5", ANTHROPIC, "anthropic", 867000, 128000, E5, true],
63
- ["claude-opus-4-8", "Opus 4.8", ANTHROPIC, "anthropic", 867000, 128000, E5, true],
64
- ["claude-sonnet-5-5", "Sonnet 5.5", ANTHROPIC, "anthropic", 872000, 128000, E5, true],
65
- ["claude-sonnet-5", "Sonnet 5", ANTHROPIC, "anthropic", 872000, 128000, E5, true],
66
- ["claude-sonnet-4-6", "Sonnet 4.6", ANTHROPIC, "anthropic", 931000, 64000, ["low", "medium", "high", "max"], true],
67
- ["claude-haiku-5-5", "Haiku 5.5", ANTHROPIC, "anthropic", 872000, 128000, E5, true],
68
- ["claude-haiku-4-5-20251001", "Haiku 4.5", ANTHROPIC, "anthropic", 0, 0, ["low", "medium", "high"], true],
69
- ["gpt-6.1-sol", "GPT-6.1 Sol", RESPONSES, "openai", 1050000, 128000, E5, true],
70
- ["gpt-6-sol", "GPT-6 Sol", RESPONSES, "openai", 1050000, 128000, E6, true],
71
- ["gpt-6-astra", "GPT-6 Astra", RESPONSES, "openai", 1050000, 128000, E5, true],
72
- ["gpt-6-luna", "GPT-6 Luna", RESPONSES, "openai", 1050000, 128000, E6, true],
73
- ["gpt-5.6-sol", "GPT-5.6 Sol", RESPONSES, "openai", 1050000, 128000, E6, true],
74
- ["gpt-5.6-terra", "GPT-5.6 Terra", RESPONSES, "openai", 1050000, 128000, E6, true],
75
- ["gpt-5.6-luna", "GPT-5.6 Luna", RESPONSES, "openai", 1050000, 128000, E6, true],
76
- ["gpt-5.5", "GPT-5.5", RESPONSES, "openai", 1050000, 128000, E4, true],
77
- ["gpt-5.4", "GPT-5.4", RESPONSES, "openai", 1050000, 128000, E4, true],
78
- ["gpt-5.3-codex", "GPT-5.3-Codex", RESPONSES, "openai", 400000, 128000, E4, true],
79
- ["grok-4.7", "Grok 4.7", RESPONSES, "xai", 500000, 63356, E4, true],
80
- ["grok-4.6", "Grok 4.6", RESPONSES, "xai", 200000, 63356, E4, true],
59
+ ["claude-fable-5.1", "Fable 5.1", ANTHROPIC, "anthropic", 867000, 128000, E5, true, 4],
60
+ ["claude-fable-5", "Fable 5", ANTHROPIC, "anthropic", 867000, 128000, E5, true, 4],
61
+ ["claude-opus-5-5", "Opus 5.5", ANTHROPIC, "anthropic", 872000, 128000, E5, true, 1.6],
62
+ ["claude-opus-5", "Opus 5", ANTHROPIC, "anthropic", 867000, 128000, E5, true, 2],
63
+ ["claude-opus-4-8", "Opus 4.8", ANTHROPIC, "anthropic", 867000, 128000, E5, true, 2],
64
+ ["claude-sonnet-5-5", "Sonnet 5.5", ANTHROPIC, "anthropic", 872000, 128000, E5, true, 0.8],
65
+ ["claude-sonnet-5", "Sonnet 5", ANTHROPIC, "anthropic", 872000, 128000, E5, true, 0.8],
66
+ ["claude-sonnet-4-6", "Sonnet 4.6", ANTHROPIC, "anthropic", 931000, 64000, ["low", "medium", "high", "max"], true, 1.2],
67
+ ["claude-haiku-5-5", "Haiku 5.5", ANTHROPIC, "anthropic", 872000, 128000, E5, true, 0.04],
68
+ ["claude-haiku-4-5-20251001", "Haiku 4.5", ANTHROPIC, "anthropic", 0, 0, ["low", "medium", "high"], true, 0.4],
69
+ ["gpt-6.1-sol", "GPT-6.1 Sol", RESPONSES, "openai", 1050000, 128000, E5, true, 0.8],
70
+ ["gpt-6-sol", "GPT-6 Sol", RESPONSES, "openai", 1050000, 128000, E6, true, 0.8],
71
+ ["gpt-6-astra", "GPT-6 Astra", RESPONSES, "openai", 1050000, 128000, E5, true, 4],
72
+ ["gpt-6-luna", "GPT-6 Luna", RESPONSES, "openai", 1050000, 128000, E6, true, 0.04],
73
+ ["gpt-5.6-sol", "GPT-5.6 Sol", RESPONSES, "openai", 1050000, 128000, E6, true, 2],
74
+ ["gpt-5.6-terra", "GPT-5.6 Terra", RESPONSES, "openai", 1050000, 128000, E6, true, 0.8],
75
+ ["gpt-5.6-luna", "GPT-5.6 Luna", RESPONSES, "openai", 1050000, 128000, E6, true, 0.08],
76
+ ["gpt-5.5", "GPT-5.5", RESPONSES, "openai", 1050000, 128000, E4, true, 2],
77
+ ["gpt-5.4", "GPT-5.4", RESPONSES, "openai", 1050000, 128000, E4, true, 1],
78
+ ["gpt-5.3-codex", "GPT-5.3-Codex", RESPONSES, "openai", 400000, 128000, E4, true, 0.7],
79
+ ["grok-4.7", "Grok 4.7", RESPONSES, "xai", 500000, 63356, E4, true, 0.8],
80
+ ["grok-4.6", "Grok 4.6", RESPONSES, "xai", 200000, 63356, E4, true, 0.8],
81
81
  // reasoning is mandatory on GLM-5.3 and GLM-5.3-Flash (Fireworks turns
82
82
  // none away): low is their least. GLM-5.2, Kimi K3 and DeepSeek V4.1
83
83
  // Flash stop thinking at none, droid's "off" (yetone/magpie#899).
84
- ["glm-5.3", "GLM-5.3", CHAT, "fireworks", 1040000, 131072, ["low", "high", "max"], false],
85
- ["glm-5.3-flash", "GLM-5.3-Flash", CHAT, "fireworks", 1048576, 131072, ["low", "high", "max"], true],
86
- ["glm-5.2", "GLM-5.2", CHAT, "baseten", 1040000, 131072, ["none", "high", "max"], false],
87
- ["kimi-k3", "Kimi K3", CHAT, "fireworks", 262144, 65536, ["none", "low", "high", "max"], true],
88
- ["deepseek-v4.1-flash", "DeepSeek V4.1 Flash", CHAT, "fireworks", 1040000, 131072, ["none", "low", "high", "max"], true],
89
- ["qwen3.8-max", "Qwen3.8 Max", CHAT, "fireworks", 262144, 131072, ["low", "medium", "xhigh"], false],
90
- ["minimax-m3", "MiniMax M3", CHAT, "fireworks", 512000, 64000, ["high"], true],
91
- ["minimax-m2.7", "MiniMax M2.7", ANTHROPIC, "fireworks", 196600, 64000, ["high"], false],
92
- ["mistral-large-4", "Mistral Large 4", CHAT, "mistral", 524288, 64000, ["high"], true],
93
- ["mistral-medium-3.5", "Mistral Medium 3.5", CHAT, "mistral", 256000, 64000, ["high"], true],
94
- ["nemotron-3-ultra", "Nemotron 3 Ultra", CHAT, "baseten", 202000, 65536, ["high"], false],
95
- ["gemini-4-argon", "Gemini 4 Argon", GEMINI, "google", 1000000, 65536, ["low", "medium", "high"], true],
96
- ["gemini-3.1-pro-preview", "Gemini 3.1 Pro", GEMINI, "google", 1000000, 65536, ["low", "medium", "high"], true],
97
- ["gemini-3.8-flash", "Gemini 3.8 Flash", GEMINI, "google", 1000000, 65536, ["low", "medium", "high"], true],
98
- ["gemini-3.7-flash", "Gemini 3.7 Flash", GEMINI, "google", 1000000, 65536, ["low", "medium", "high"], true],
99
- ["gemini-3.6-flash", "Gemini 3.6 Flash", GEMINI, "google", 1000000, 65536, ["low", "medium", "high"], true],
100
- ["gemini-3.5-flash", "Gemini 3.5 Flash", GEMINI, "google", 1000000, 65536, ["minimal", "low", "medium", "high"], true],
101
- ["gemini-3-flash-preview", "Gemini 3 Flash", GEMINI, "google", 1000000, 65536, ["minimal", "low", "medium", "high"], true],
84
+ ["glm-5.3", "GLM-5.3", CHAT, "fireworks", 1040000, 131072, ["low", "high", "max"], false, 0.56],
85
+ ["glm-5.3-flash", "GLM-5.3-Flash", CHAT, "fireworks", 1048576, 131072, ["low", "high", "max"], true, 0.06],
86
+ ["glm-5.2", "GLM-5.2", CHAT, "baseten", 1040000, 131072, ["none", "high", "max"], false, 0.56],
87
+ ["kimi-k3", "Kimi K3", CHAT, "fireworks", 262144, 65536, ["none", "low", "high", "max"], true, 1.2],
88
+ ["deepseek-v4.1-flash", "DeepSeek V4.1 Flash", CHAT, "fireworks", 1040000, 131072, ["none", "low", "high", "max"], true, 0.12],
89
+ ["qwen3.8-max", "Qwen3.8 Max", CHAT, "fireworks", 262144, 131072, ["low", "medium", "xhigh"], false, 0.8],
90
+ ["minimax-m3", "MiniMax M3", CHAT, "fireworks", 512000, 64000, ["high"], true, 0.12],
91
+ ["minimax-m2.7", "MiniMax M2.7", ANTHROPIC, "fireworks", 196600, 64000, ["high"], false, 0.12],
92
+ ["mistral-large-4", "Mistral Large 4", CHAT, "mistral", 524288, 64000, ["high"], true, 0.544],
93
+ ["mistral-medium-3.5", "Mistral Medium 3.5", CHAT, "mistral", 256000, 64000, ["high"], true, 0.6],
94
+ ["nemotron-3-ultra", "Nemotron 3 Ultra", CHAT, "baseten", 202000, 65536, ["high"], false, 0.24],
95
+ ["gemini-4-argon", "Gemini 4 Argon", GEMINI, "google", 1000000, 65536, ["low", "medium", "high"], true, 1.6],
96
+ ["gemini-3.1-pro-preview", "Gemini 3.1 Pro", GEMINI, "google", 1000000, 65536, ["low", "medium", "high"], true, 0.8],
97
+ ["gemini-3.8-flash", "Gemini 3.8 Flash", GEMINI, "google", 1000000, 65536, ["low", "medium", "high"], true, 0.6],
98
+ ["gemini-3.7-flash", "Gemini 3.7 Flash", GEMINI, "google", 1000000, 65536, ["low", "medium", "high"], true, 0.6],
99
+ ["gemini-3.6-flash", "Gemini 3.6 Flash", GEMINI, "google", 1000000, 65536, ["low", "medium", "high"], true, 0.6],
100
+ ["gemini-3.5-flash", "Gemini 3.5 Flash", GEMINI, "google", 1000000, 65536, ["minimal", "low", "medium", "high"], true, 0.6],
101
+ ["gemini-3-flash-preview", "Gemini 3 Flash", GEMINI, "google", 1000000, 65536, ["minimal", "low", "medium", "high"], true, 0.2],
102
102
  // The Fast Mode models come last: magpie shows a provider with no models
103
103
  // picked only the first 24 of its list, and placed beside their base
104
104
  // models in 0.1.27 they pushed GPT-5.3-Codex, Grok 4.7, Grok 4.6, GLM-5.3
105
105
  // and GLM-5.3-Flash out of it (#119, psylch). A Fast Mode model costs more credits, so it is the one
106
106
  // to pick; asking its base model for fast reaches it all the same
107
107
  // (fastBody).
108
- ["claude-opus-5-5-fast", "Opus 5.5 Fast Mode", ANTHROPIC, "anthropic", 872000, 128000, E5, true],
109
- ["claude-opus-5-fast", "Opus 5 Fast Mode", ANTHROPIC, "anthropic", 867000, 128000, E5, true],
110
- ["claude-opus-4-8-fast", "Opus 4.8 Fast Mode", ANTHROPIC, "anthropic", 867000, 128000, E5, true],
111
- ["gpt-5.5-fast", "GPT-5.5 Fast Mode", RESPONSES, "openai", 1050000, 128000, E4, true],
112
- ["gpt-5.4-fast", "GPT-5.4 Fast Mode", RESPONSES, "openai", 1050000, 128000, E4, true],
113
- ["gpt-5.3-codex-fast", "GPT-5.3-Codex Fast Mode", RESPONSES, "openai", 400000, 128000, E4, true],
114
- ].map(([id, name, wire, upstream, context, output, efforts, images]) => ({ id, name, wire, upstream, context, output, efforts, images }))
108
+ ["claude-opus-5-5-fast", "Opus 5.5 Fast Mode", ANTHROPIC, "anthropic", 872000, 128000, E5, true, 3.2],
109
+ ["claude-opus-5-fast", "Opus 5 Fast Mode", ANTHROPIC, "anthropic", 867000, 128000, E5, true, 4],
110
+ ["claude-opus-4-8-fast", "Opus 4.8 Fast Mode", ANTHROPIC, "anthropic", 867000, 128000, E5, true, 4],
111
+ ["gpt-5.5-fast", "GPT-5.5 Fast Mode", RESPONSES, "openai", 1050000, 128000, E4, true, 5],
112
+ ["gpt-5.4-fast", "GPT-5.4 Fast Mode", RESPONSES, "openai", 1050000, 128000, E4, true, 2],
113
+ ["gpt-5.3-codex-fast", "GPT-5.3-Codex Fast Mode", RESPONSES, "openai", 400000, 128000, E4, true, 1.4],
114
+ ].map(([id, name, wire, upstream, context, output, efforts, images, rate]) => ({ id, name, wire, upstream, context, output, efforts, images, rate }))
115
115
 
116
116
  const byId = new Map(MODELS.map((m) => [m.id, m]))
117
117
 
@@ -162,9 +162,35 @@ function variant(npm, effort) {
162
162
  return { reasoningEffort: effort }
163
163
  }
164
164
 
165
- function configModels() {
165
+ // Factory's running promotions, as droid's registry keeps them
166
+ // (cost.promotions): while one runs, a model's price is its token multiplier
167
+ // less the promotion's discount. magpie shows the price now with the price
168
+ // before it struck through (rateWas), as it does for WorkBuddy's credits and
169
+ // Qoder's price factors, so a cheap model is picked without opening droid. A
170
+ // promotion that has ended is no discount at all.
171
+ const PROMOTIONS = [
172
+ { id: "gpt-5.6-sol", discount: 0.2, from: "2026-08-22T00:00:00Z", to: "2026-11-22T00:00:00Z" },
173
+ { id: "mistral-large-4", discount: 0.5, from: "2026-10-06T00:00:00Z", to: "2026-10-20T00:00:00Z" },
174
+ { id: "gemini-4-argon", discount: 0.5, from: "2026-10-13T16:00:00Z", to: "2026-12-01T08:00:00Z" },
175
+ { id: "gemini-3.8-flash", discount: 0.5, from: "2026-08-17T21:10:19Z", to: "2027-01-01T08:00:00Z" },
176
+ { id: "gemini-3.7-flash", discount: 0.5, from: "2026-08-17T21:10:19Z", to: "2027-01-01T08:00:00Z" },
177
+ ]
178
+
179
+ // rateOf is a model's price as magpie keeps it: rate, the token multiplier a
180
+ // request costs now, and rateWas, the multiplier before a promotion running
181
+ // now (0: none). droid reads it the same way (ib(cost, now)): the first
182
+ // promotion that has started and not ended takes its discount off the
183
+ // multiplier.
184
+ function rateOf(m, now = Date.now()) {
185
+ const p = PROMOTIONS.find((p) => p.id === m.id && now >= Date.parse(p.from) && now < Date.parse(p.to))
186
+ if (!p) return { rate: m.rate, rateWas: 0 }
187
+ return { rate: Number((m.rate * (1 - p.discount)).toFixed(4)), rateWas: m.rate }
188
+ }
189
+
190
+ function configModels(now = Date.now()) {
166
191
  const out = {}
167
192
  for (const m of MODELS) {
193
+ const { rate, rateWas } = rateOf(m, now)
168
194
  out[m.id] = {
169
195
  id: m.id,
170
196
  name: m.name,
@@ -172,16 +198,62 @@ function configModels() {
172
198
  reasoning: m.efforts.length > 0,
173
199
  attachment: m.images,
174
200
  tool_call: true,
175
- temperature: true,
201
+ // OpenAI's reasoning models take no temperature (droidBody)
202
+ temperature: m.upstream !== "openai",
176
203
  modalities: { input: m.images ? ["text", "image"] : ["text"], output: ["text"] },
177
204
  cost: { input: 0, output: 0 },
178
205
  limit: { context: m.context, output: m.output },
206
+ // magpie's own: the token multiplier a request costs now, and the
207
+ // multiplier before a promotion running now (0: none)
208
+ rate,
209
+ rateWas,
179
210
  variants: Object.fromEntries(m.efforts.map((e) => [e, variant(m.wire.npm, e)])),
180
211
  }
181
212
  }
182
213
  return out
183
214
  }
184
215
 
216
+ // providerModel is a model as magpie's provider hook keeps it: the same model
217
+ // configModels gives OpenCode, in magpie's shape. magpie reads a provider's
218
+ // models from this hook, and a config hook's models without the price beside
219
+ // them, so the token multiplier droid shows a model by (rateOf) is carried
220
+ // here.
221
+ function providerModel(provider, m, now = Date.now()) {
222
+ const { rate, rateWas } = rateOf(m, now)
223
+ return {
224
+ id: m.id,
225
+ providerID: provider.id,
226
+ name: m.name,
227
+ api: { id: m.id, url: m.wire.api, npm: m.wire.npm },
228
+ status: "active",
229
+ headers: {},
230
+ options: {},
231
+ cost: { input: 0, output: 0, cache: { read: 0, write: 0 } },
232
+ limit: { context: m.context, output: m.output },
233
+ capabilities: {
234
+ temperature: true,
235
+ reasoning: m.efforts.length > 0,
236
+ attachment: m.images,
237
+ toolcall: true,
238
+ input: { text: true, image: m.images, audio: false, video: false, pdf: false },
239
+ output: { text: true, image: false, audio: false, video: false, pdf: false },
240
+ interleaved: false,
241
+ },
242
+ release_date: "",
243
+ variants: Object.fromEntries(m.efforts.map((e) => [e, variant(m.wire.npm, e)])),
244
+ free: false,
245
+ // magpie's own: the token multiplier a request costs now, and the
246
+ // multiplier before a promotion running now (0: none)
247
+ rate,
248
+ rateWas,
249
+ }
250
+ }
251
+
252
+ // providerModels is the list the provider hook answers with, at now.
253
+ function providerModels(provider, now = Date.now()) {
254
+ return Object.fromEntries(MODELS.map((m) => [m.id, providerModel(provider, m, now)]))
255
+ }
256
+
185
257
  // ---- tokens -----------------------------------------------------------------
186
258
 
187
259
  class FactoryStatus extends Error {
@@ -474,11 +546,18 @@ function droidBody(path, body) {
474
546
  if (path.endsWith("/responses")) {
475
547
  const v = m.instructions
476
548
  if (v !== undefined && v !== null && typeof v !== "string") return body // not something droid sends
549
+ // OpenAI's reasoning models turn temperature and top_p away with a 400,
550
+ // and droid sends neither (Nowledge Mem on gpt-6.1-sol, magpie's Discord)
551
+ const sampled = byId.get(m.model)?.upstream === "openai" && ("temperature" in m || "top_p" in m)
552
+ if (sampled) {
553
+ delete m.temperature
554
+ delete m.top_p
555
+ }
477
556
  const opened = promptOpenings(m.input)
478
557
  const ins = promptText(v ?? "")
479
558
  if (!ins.startsWith(DROID_LINE)) m.instructions = ins.trim() === "" ? DROID_LINE : DROID_LINE + "\n" + ins
480
559
  else if (ins !== v || opened) m.instructions = ins
481
- else return body
560
+ else if (!sampled) return body
482
561
  } else if (path.endsWith("/chat/completions")) {
483
562
  if (!Array.isArray(m.messages)) return body
484
563
  const opened = promptOpenings(m.messages, true)
@@ -1613,6 +1692,15 @@ export const FactoryAuthPlugin = async ({ client }) => {
1613
1692
  }
1614
1693
 
1615
1694
  return {
1695
+ // magpie's list for this provider: the same models configModels gives
1696
+ // OpenCode, each with the token multiplier droid shows it by, which
1697
+ // magpie shows beside the model as its rate
1698
+ provider: {
1699
+ id: PROVIDER,
1700
+ async models(provider) {
1701
+ return providerModels(provider)
1702
+ },
1703
+ },
1616
1704
  config: async (config) => {
1617
1705
  config.provider ??= {}
1618
1706
  const was = config.provider[PROVIDER] ?? {}
@@ -1778,4 +1866,4 @@ export const FactoryAuthPlugin = async ({ client }) => {
1778
1866
  }
1779
1867
 
1780
1868
  // for tests
1781
- export const _internal = { fastBody, FAST_BETA, limitWindows, windowEnd, dollars, vendorError, CORE, droidBody, DROID_LINE, geminiBody, leaveOutAttachments, LEFT_OUT, TOO_LARGE_HINT }
1869
+ export const _internal = { fastBody, FAST_BETA, limitWindows, windowEnd, dollars, vendorError, CORE, droidBody, DROID_LINE, geminiBody, leaveOutAttachments, LEFT_OUT, TOO_LARGE_HINT, MODELS, PROMOTIONS, rateOf, configModels, providerModel, providerModels }
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@magpie-community/opencode-factory-auth",
3
- "version": "0.1.30",
3
+ "version": "0.1.32",
4
4
  "description": "Sign in to a Factory (Droid) subscription from OpenCode or magpie",
5
5
  "type": "module",
6
6
  "main": "./index.mjs",