roadmodel 0.2.64__tar.gz → 0.2.67__tar.gz

{roadmodel-0.2.64 → roadmodel-0.2.67}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: roadmodel
-Version: 0.2.64
+Version: 0.2.67
 Summary: BYO-key CLI that recommends the right AI model, platform, and settings for a prompt.
 Project-URL: Homepage, https://roadmodel.ai
 Project-URL: Repository, https://github.com/nathanramoscfa/roadmodel

{roadmodel-0.2.64 → roadmodel-0.2.67}/docs/benchmarks.json

@@ -6,7 +6,7 @@
     "endpoint": "https://artificialanalysis.ai/api/v2/data/llms/models",
     "attribution": "Benchmark data © Artificial Analysis, used under its API terms with attribution."
   },
-  "generated_at_utc": "2026-10-01T22:10:29Z",
+  "generated_at_utc": "2026-10-05T22:11:00Z",
   "model_count": 54,
   "unmapped": [],
   "missing_slugs": [],
@@ -38,13 +38,13 @@
         "tau2": 0.546783625730994,
         "tau_banking": 0.0927835051546392
       },
-      "median_output_tokens_per_second": 125.146,
-      "median_time_to_first_token_seconds": 13.494
+      "median_output_tokens_per_second": 107.429,
+      "median_time_to_first_token_seconds": 7.79
     },
     "claude-fable-5": {
       "aa_id": "cd55210d-358e-4df1-ba9c-9acb5f186cc9",
       "aa_slug": "claude-fable-5",
-      "aa_name": "Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback)",
+      "aa_name": "Claude Fable 5 (Max, Opus 4.8 Fallback)",
       "aa_creator": "Anthropic",
       "release_date": "2026-06-09",
       "evaluations": {
@@ -73,7 +73,7 @@
     "claude-fable-5.1": {
       "aa_id": "3e87c73e-a257-495e-9730-367a66229811",
       "aa_slug": "claude-fable-5-1",
-      "aa_name": "Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)",
+      "aa_name": "Claude Fable 5.1 (Max, Default Fallback)",
       "aa_creator": "Anthropic",
       "release_date": "2026-09-01",
       "evaluations": {
@@ -96,13 +96,13 @@
         "tau2": null,
         "tau_banking": 0.472164948453608
       },
-      "median_output_tokens_per_second": 68.681,
-      "median_time_to_first_token_seconds": 127.394
+      "median_output_tokens_per_second": 68.612,
+      "median_time_to_first_token_seconds": 145.612
     },
     "claude-opus-5": {
       "aa_id": "b8fc61f7-5e9a-49e6-8547-6ac56db24627",
       "aa_slug": "claude-opus-5",
-      "aa_name": "Claude Opus 5 (Adaptive Reasoning, Max Effort)",
+      "aa_name": "Claude Opus 5 (Max)",
       "aa_creator": "Anthropic",
       "release_date": "2026-07-24",
       "evaluations": {
@@ -131,7 +131,7 @@
     "claude-opus-5-5": {
       "aa_id": "2f3c4dc9-a450-4303-8697-0237257cf08f",
       "aa_slug": "claude-opus-5-5",
-      "aa_name": "Claude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)",
+      "aa_name": "Claude Opus 5.5 (Max, Default Fallback)",
       "aa_creator": "Anthropic",
       "release_date": "2026-09-22",
       "evaluations": {
@@ -154,13 +154,13 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 96.458,
-      "median_time_to_first_token_seconds": 477.23
+      "median_output_tokens_per_second": 96.593,
+      "median_time_to_first_token_seconds": 489.353
     },
     "claude-sonnet-5": {
       "aa_id": "23c86e4a-c769-43c0-a056-79e3cd15834f",
       "aa_slug": "claude-sonnet-5",
-      "aa_name": "Claude Sonnet 5 (Adaptive Reasoning, Max Effort)",
+      "aa_name": "Claude Sonnet 5 (Max)",
       "aa_creator": "Anthropic",
       "release_date": "2026-06-30",
       "evaluations": {
@@ -183,13 +183,13 @@
         "tau2": null,
         "tau_banking": 0.37319587628866
       },
-      "median_output_tokens_per_second": 94.143,
-      "median_time_to_first_token_seconds": 162.437
+      "median_output_tokens_per_second": 0,
+      "median_time_to_first_token_seconds": 0
     },
     "claude-sonnet-5-5": {
       "aa_id": "b171d979-5ec1-45de-b8b9-db1ee65e77ec",
       "aa_slug": "claude-sonnet-5-5",
-      "aa_name": "Claude Sonnet 5.5 (Adaptive Reasoning, Max Effort, Default Fallback)",
+      "aa_name": "Claude Sonnet 5.5 (Max, Default Fallback)",
       "aa_creator": "Anthropic",
       "release_date": "2026-09-28",
       "evaluations": {
@@ -212,8 +212,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 145.124,
-      "median_time_to_first_token_seconds": 276.058
+      "median_output_tokens_per_second": 137.507,
+      "median_time_to_first_token_seconds": 269.765
     },
     "deepseek-flash": {
       "aa_id": "dbe7c625-3100-4463-b479-a228c41f75dd",
@@ -241,8 +241,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 226.5,
-      "median_time_to_first_token_seconds": 0.844
+      "median_output_tokens_per_second": 214.176,
+      "median_time_to_first_token_seconds": 0.802
     },
     "deepseek-v4-pro": {
       "aa_id": "83173329-c09a-41f1-a028-a282a5f908d7",
@@ -270,8 +270,8 @@
         "tau2": null,
         "tau_banking": 0.395876288659794
       },
-      "median_output_tokens_per_second": 101.435,
-      "median_time_to_first_token_seconds": 1.156
+      "median_output_tokens_per_second": 103.076,
+      "median_time_to_first_token_seconds": 1.112
     },
     "gemini-2.5-flash": {
       "aa_id": "219ed587-60c5-4a48-9517-8480e08d0ca1",
@@ -386,8 +386,8 @@
         "tau2": 0.956140350877193,
         "tau_banking": 0.214432989690722
       },
-      "median_output_tokens_per_second": 127.145,
-      "median_time_to_first_token_seconds": 26.91
+      "median_output_tokens_per_second": 123.821,
+      "median_time_to_first_token_seconds": 15.32
     },
     "gemini-3.5-flash": {
       "aa_id": "0097ebf5-124f-42f6-9463-33b00e711f03",
@@ -502,8 +502,8 @@
         "tau2": null,
         "tau_banking": 0.449484536082474
       },
-      "median_output_tokens_per_second": 244.009,
-      "median_time_to_first_token_seconds": 13.552
+      "median_output_tokens_per_second": 223.213,
+      "median_time_to_first_token_seconds": 9.822
     },
     "glm-4.5-air": {
       "aa_id": "5d303dc9-c027-401f-9803-4e9aa3331007",
@@ -618,8 +618,8 @@
         "tau2": null,
         "tau_banking": 0.503092783505155
       },
-      "median_output_tokens_per_second": 69.965,
-      "median_time_to_first_token_seconds": 2.71
+      "median_output_tokens_per_second": 75.232,
+      "median_time_to_first_token_seconds": 3.1
     },
     "glm-5.3-flash": {
       "aa_id": "19496b81-9f41-4214-a77a-1df803b3c5ae",
@@ -647,8 +647,8 @@
         "tau2": null,
         "tau_banking": 0.472164948453608
       },
-      "median_output_tokens_per_second": 49.583,
-      "median_time_to_first_token_seconds": 3.195
+      "median_output_tokens_per_second": 49.967,
+      "median_time_to_first_token_seconds": 2.638
     },
     "gpt-5": {
       "aa_id": "48e50f00-1fd1-4acc-b337-61078aa341e6",
@@ -850,8 +850,8 @@
         "tau2": 0.859649122807018,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 96.414,
-      "median_time_to_first_token_seconds": 54.486
+      "median_output_tokens_per_second": 102.301,
+      "median_time_to_first_token_seconds": 34.347
     },
     "gpt-5.4": {
       "aa_id": "a89c4b28-2d8c-456e-88ea-255fb51fd2b6",
@@ -1053,8 +1053,8 @@
         "tau2": 0.862573099415205,
         "tau_banking": 0.402061855670103
       },
-      "median_output_tokens_per_second": 109.104,
-      "median_time_to_first_token_seconds": 115.347
+      "median_output_tokens_per_second": 116.052,
+      "median_time_to_first_token_seconds": 82.187
     },
     "gpt-6-astra": {
       "aa_id": "2f339a97-9a0d-499a-9cb5-e0db665bfa25",
@@ -1082,8 +1082,8 @@
         "tau2": null,
         "tau_banking": 0.414432989690722
       },
-      "median_output_tokens_per_second": 55.68,
-      "median_time_to_first_token_seconds": 192.954
+      "median_output_tokens_per_second": 64.764,
+      "median_time_to_first_token_seconds": 191.318
     },
     "gpt-6-luna": {
       "aa_id": "6fb13851-40ce-4bda-ad3f-6b38e0c5daa7",
@@ -1111,8 +1111,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 139.721,
-      "median_time_to_first_token_seconds": 84.615
+      "median_output_tokens_per_second": 135.654,
+      "median_time_to_first_token_seconds": 75.426
     },
     "gpt-6-sol": {
       "aa_id": "c50ea08c-88c0-4eb7-85b8-27f2bbf6d527",
@@ -1140,8 +1140,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 90.088,
-      "median_time_to_first_token_seconds": 107.871
+      "median_output_tokens_per_second": 0,
+      "median_time_to_first_token_seconds": 0
     },
     "gpt-6.1-sol": {
       "aa_id": "b25ac058-1e33-4c31-bbc9-a20c3ad0717a",
@@ -1169,8 +1169,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 63.279,
-      "median_time_to_first_token_seconds": 139.092
+      "median_output_tokens_per_second": 60.455,
+      "median_time_to_first_token_seconds": 157.396
     },
     "gpt-oss-120b": {
       "aa_id": "f0083258-8646-45b8-8082-7aaf6c2ea82a",
@@ -1198,8 +1198,8 @@
         "tau2": 0.657894736842105,
         "tau_banking": 0.127835051546392
       },
-      "median_output_tokens_per_second": 154.167,
-      "median_time_to_first_token_seconds": 0.54
+      "median_output_tokens_per_second": 199.818,
+      "median_time_to_first_token_seconds": 0.501
     },
     "gpt-oss-20b": {
       "aa_id": "36f73aaf-d38a-4b56-a2b3-d04d17186910",
@@ -1227,8 +1227,8 @@
         "tau2": 0.60233918128655,
         "tau_banking": 0.0701030927835052
       },
-      "median_output_tokens_per_second": 192.538,
-      "median_time_to_first_token_seconds": 0.415
+      "median_output_tokens_per_second": 227.873,
+      "median_time_to_first_token_seconds": 0.427
     },
     "grok-4.3": {
       "aa_id": "948892b5-db03-4118-a4a8-ccd51ed871ea",
@@ -1314,8 +1314,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 84.78,
-      "median_time_to_first_token_seconds": 19.787
+      "median_output_tokens_per_second": 72.511,
+      "median_time_to_first_token_seconds": 9.66
     },
     "kimi-k2.7-code": {
       "aa_id": "8d0cb231-7303-452c-9923-a9620b948475",
@@ -1343,8 +1343,8 @@
         "tau2": 0.900584795321637,
         "tau_banking": 0.202061855670103
       },
-      "median_output_tokens_per_second": 77.467,
-      "median_time_to_first_token_seconds": 1.204
+      "median_output_tokens_per_second": 105.681,
+      "median_time_to_first_token_seconds": 1.142
     },
     "kimi-k3": {
       "aa_id": "f7d2fc3e-1f7b-405f-818c-07952a4af78f",
@@ -1372,8 +1372,8 @@
         "tau2": null,
         "tau_banking": 0.45979381443299
       },
-      "median_output_tokens_per_second": 37.907,
-      "median_time_to_first_token_seconds": 3.013
+      "median_output_tokens_per_second": 45.199,
+      "median_time_to_first_token_seconds": 2.455
     },
     "mistral-large-3": {
       "aa_id": "4928e950-7f37-4475-b0dc-c5bad781a321",
@@ -1401,8 +1401,8 @@
         "tau2": 0.245614035087719,
         "tau_banking": 0.0577319587628866
       },
-      "median_output_tokens_per_second": 75.045,
-      "median_time_to_first_token_seconds": 0.721
+      "median_output_tokens_per_second": 79.265,
+      "median_time_to_first_token_seconds": 0.692
     },
     "mistral-medium-3.5": {
       "aa_id": "dd059b25-d82a-4ead-82a4-4adceaaec48b",
@@ -1430,8 +1430,8 @@
         "tau2": 0.941520467836257,
         "tau_banking": 0.150515463917526
       },
-      "median_output_tokens_per_second": 167.324,
-      "median_time_to_first_token_seconds": 0.625
+      "median_output_tokens_per_second": 168.037,
+      "median_time_to_first_token_seconds": 0.6
     },
     "mistral-small-4": {
       "aa_id": "3fd96175-4ef1-434c-8795-f873aec2abc1",
@@ -1459,8 +1459,8 @@
         "tau2": 0.412280701754386,
         "tau_banking": 0.0494845360824742
       },
-      "median_output_tokens_per_second": 173.192,
-      "median_time_to_first_token_seconds": 0.626
+      "median_output_tokens_per_second": 171.598,
+      "median_time_to_first_token_seconds": 0.559
     },
     "muse-spark-1.3": {
       "aa_id": "9999672c-b687-4026-9c15-5cef02ff53bb",
@@ -1488,13 +1488,13 @@
         "tau2": null,
         "tau_banking": 0.505154639175258
       },
-      "median_output_tokens_per_second": 171.012,
-      "median_time_to_first_token_seconds": 37.827
+      "median_output_tokens_per_second": 122.652,
+      "median_time_to_first_token_seconds": 28.442
     },
     "opus-4.7": {
       "aa_id": "e9a09db3-8fd6-41dd-ba2f-20e0a2bff7f2",
       "aa_slug": "claude-opus-4-7",
-      "aa_name": "Claude Opus 4.7 (Adaptive Reasoning, Max Effort)",
+      "aa_name": "Claude Opus 4.7 (Max)",
       "aa_creator": "Anthropic",
       "release_date": "2026-04-16",
       "evaluations": {
@@ -1523,7 +1523,7 @@
     "opus-4.8": {
       "aa_id": "992b7b84-5069-4c6a-9295-834252553d50",
       "aa_slug": "claude-opus-4-8",
-      "aa_name": "Claude Opus 4.8 (Adaptive Reasoning, Max Effort)",
+      "aa_name": "Claude Opus 4.8 (Max)",
       "aa_creator": "Anthropic",
       "release_date": "2026-05-28",
       "evaluations": {
@@ -1552,7 +1552,7 @@
     "sonnet-4.6": {
       "aa_id": "df8d14e0-3997-4e4d-b4ad-9c047acc9c69",
       "aa_slug": "claude-sonnet-4-6-adaptive",
-      "aa_name": "Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort)",
+      "aa_name": "Claude Sonnet 4.6 (Max)",
       "aa_creator": "Anthropic",
       "release_date": "2026-02-17",
       "evaluations": {

{roadmodel-0.2.64 → roadmodel-0.2.67}/docs/catalog.json

@@ -1,9 +1,9 @@
 {
   "schema_version": "4",
   "output_contract_version": 2,
-  "generated_at_utc": "2026-10-02T18:08:21Z",
+  "generated_at_utc": "2026-10-06T20:14:04Z",
   "source_doc_sha256": {
-    "model-selector.txt": "81c5530a28521df18ebe14810e56cd72e65f1e30ad93d0522d31dbf362516e93",
+    "model-selector.txt": "b7abeb92fb4d9b85b70a4544369a49de103d5695bf6dd77ea996dc4e85db0a31",
     "model-tier-cost-scale.md": "2b3b0eb82237eb5cb5bf468a3fe971bcaf0160ad7d0bc700698d2c16640bc41d"
   },
   "models": [
@@ -1424,7 +1424,7 @@
         "speed": "B"
       },
       "jurisdiction": "us",
-      "headline_benchmarks": "AA Intelligence Index 30.1 (adaptive reasoning, max effort); HLE 33.6%; SciCode 50.1; Terminal-Bench 4.0 3.0; AA-LCR 0.800; LMArena Text Elo 1457.7 (#41); AA-Omniscience 12.4",
+      "headline_benchmarks": "AA Intelligence Index 30.1 (max); HLE 33.6%; SciCode 50.1; Terminal-Bench 4.0 3.0; AA-LCR 0.800; LMArena Text Elo 1457.7 (#41); AA-Omniscience 12.4",
       "pricing_notes": "Hidden by default; Requires Max Mode on legacy request-based plans; Up to 1M tokens with extended context at the same per-token rates (no long-context surcharge)",
       "best_for": "Strong multimodal input and long context at a high price. Its AA Intelligence Index of 30.1 ranks 33rd of 54 measured models.",
       "superseded_by": null,
@@ -2048,7 +2048,7 @@
       "exposes_max_mode": "no",
       "exposes_thinking": "yes",
       "exposes_orchestration": "no",
-      "best_for": "Per-token AGGREGATOR that resells other makers' models behind one OpenAI-format endpoint and one key (openrouter.ai) \u2014 like Cursor's pool but paid per call, not a subscription. Reaches most of the catalog: supports-models is a POINT-IN-TIME snapshot of openrouter.ai/models at authoring (refreshed 2026-10-02; absent: composer-2.5 which is Cursor-only, gemini-3-pro which OpenRouter serves only as an image endpoint, mistral-large-3 which it serves only as a batch endpoint) that the operator should trust LESS than a provider-direct method \u2014 prefer the maker's own method when the operator holds that key. BILLING: OpenRouter charges the maker's list price plus its platform fee (~5% on credit purchases at authoring), so the catalog price is a FLOOR for this method, never an exact estimate. ROUTING: a request may be served by any of OpenRouter's upstream hosts for that model; the MODEL-level jurisdiction filter (Step 0b) still governs which models are eligible, and this method's `us` code is OpenRouter's own (the operator's counterparty), NOT the upstream host's. Exposes the maker's reasoning dial where the routed model documents one (`reasoning_effort` on OpenAI / gpt-oss / GLM / DeepSeek families). Never the maker of any model it resells: the cross-provider BACKUP guard resolves makers through provider-direct methods, exactly as it does for Cursor."
+      "best_for": "Per-token AGGREGATOR that resells other makers' models behind one OpenAI-format endpoint and one key (openrouter.ai) \u2014 like Cursor's pool but paid per call, not a subscription. Reaches most of the catalog: supports-models is synced daily from openrouter.ai/models (OpenRouter's public model list): each catalogued model it lists as a text chat route, matched by name (absent today: composer-2.5, which is Cursor-only, and gemini-3-pro, which OpenRouter serves only as an image model). The operator should trust it LESS than a provider-direct method \u2014 prefer the maker's own method when the operator holds that key. BILLING: OpenRouter charges the maker's list price plus its platform fee (~5% on credit purchases at authoring), so the catalog price is a FLOOR for this method, never an exact estimate. ROUTING: a request may be served by any of OpenRouter's upstream hosts for that model; the MODEL-level jurisdiction filter (Step 0b) still governs which models are eligible, and this method's `us` code is OpenRouter's own (the operator's counterparty), NOT the upstream host's. Exposes the maker's reasoning dial where the routed model documents one (`reasoning_effort` on OpenAI / gpt-oss / GLM / DeepSeek families). Never the maker of any model it resells: the cross-provider BACKUP guard resolves makers through provider-direct methods, exactly as it does for Cursor."
     },
     {
       "id": "xai-api",

{roadmodel-0.2.64 → roadmodel-0.2.67}/docs/model-selector.txt

@@ -456,7 +456,19 @@
       default. It also makes Opus 4.7+ and Fable default to a 1M context
       window on Bedrock, Vertex, Foundry and the Claude apps gateway. The
       `/effort` vocabulary and the extended-thinking controls are
-      unchanged.
+      unchanged. 2.1.288 fixes resumed sessions dropping the model's
+      earlier thinking, retries thinking-only responses after a
+      mid-response API timeout, and makes a plugin-defined agent spawned
+      by name run with its own effort instead of the default. 2.1.289 is a
+      plugin / mod / permission-rule bug-fix release. Neither changes the
+      `/effort` vocabulary or the extended-thinking controls. 2.1.290
+      makes three effort/thinking fixes. A flagged message retried on a
+      fallback model with a different saved level no longer changes the
+      effort level. A reply that the output content filter stopped while
+      Claude was still thinking is now retried once. `/effort` sent from
+      `claude agents` to a busy background session now applies right
+      away. 2.1.291 is a hotfix for two regressions. Neither release
+      changes the `/effort` vocabulary or the extended-thinking controls.
     - OpenAI (Codex, OpenAI API, ChatGPT advanced controls):
       `reasoning_effort` knob — `low`, `medium`, `high`, `xhigh`, `max`,
       `ultra`. Codex's plan-mode `plan_mode_reasoning_effort` variant
@@ -1040,7 +1052,7 @@
              tier-coding="B" tier-planning="B" tier-agentic="B"
              tier-multimodal="A" tier-long-context="A" tier-knowledge="B"
              tier-speed="B"
-             headline-benchmarks="AA Intelligence Index 30.1 (adaptive reasoning, max effort); HLE 33.6%; SciCode 50.1; Terminal-Bench 4.0 3.0; AA-LCR 0.800; LMArena Text Elo 1457.7 (#41); AA-Omniscience 12.4"
+             headline-benchmarks="AA Intelligence Index 30.1 (max); HLE 33.6%; SciCode 50.1; Terminal-Bench 4.0 3.0; AA-LCR 0.800; LMArena Text Elo 1457.7 (#41); AA-Omniscience 12.4"
              pricing-notes="Hidden by default; Requires Max Mode on legacy request-based plans; Up to 1M tokens with extended context at the same per-token rates (no long-context surcharge)"
              prose-generated="best-for headline-benchmarks"
              best-for="Strong multimodal input and long context at a high price. Its AA Intelligence Index of 30.1 ranks 33rd of 54 measured models."/>
@@ -1693,7 +1705,7 @@
             supports-models="claude-sonnet-5-5,claude-opus-5-5,claude-opus-5,opus-4.8,claude-fable-5.1,claude-fable-5,opus-4.7,claude-sonnet-5,sonnet-4.6,claude-4.5-haiku,gpt-6-astra,gpt-6.1-sol,gpt-6-sol,gpt-6-luna,gpt-5.6-sol,gpt-5.5,gpt-5.6-terra,gpt-5.4,gpt-5.3-codex,gpt-5.2,gpt-5.2-codex,gpt-5.1-codex-max,gpt-5.1-codex,gpt-5.1-codex-mini,gpt-5,gpt-5.6-luna,gpt-5.4-mini,gpt-5.4-nano,gpt-5-mini,gemini-3.1-pro,gemini-3.8-flash,gemini-3.7-flash,gemini-3.6-flash,gemini-3.5-flash,gemini-3-flash,gemini-2.5-flash,grok-4.7,grok-4.6,grok-4.3,deepseek-v4-pro,deepseek-flash,mistral-medium-3.5,mistral-small-4,codestral,glm-5.3,glm-5.3-flash,glm-5.2,glm-4.6,glm-4.5-air,gpt-oss-120b,gpt-oss-20b,kimi-k3,kimi-k2.7-code,muse-spark-1.3"
             exposes-max-mode="no" exposes-thinking="yes"
             exposes-orchestration="no"
-            best-for="Per-token AGGREGATOR that resells other makers' models behind one OpenAI-format endpoint and one key (openrouter.ai) — like Cursor's pool but paid per call, not a subscription. Reaches most of the catalog: supports-models is a POINT-IN-TIME snapshot of openrouter.ai/models at authoring (refreshed 2026-10-02; absent: composer-2.5 which is Cursor-only, gemini-3-pro which OpenRouter serves only as an image endpoint, mistral-large-3 which it serves only as a batch endpoint) that the operator should trust LESS than a provider-direct method — prefer the maker's own method when the operator holds that key. BILLING: OpenRouter charges the maker's list price plus its platform fee (~5% on credit purchases at authoring), so the catalog price is a FLOOR for this method, never an exact estimate. ROUTING: a request may be served by any of OpenRouter's upstream hosts for that model; the MODEL-level jurisdiction filter (Step 0b) still governs which models are eligible, and this method's `us` code is OpenRouter's own (the operator's counterparty), NOT the upstream host's. Exposes the maker's reasoning dial where the routed model documents one (`reasoning_effort` on OpenAI / gpt-oss / GLM / DeepSeek families). Never the maker of any model it resells: the cross-provider BACKUP guard resolves makers through provider-direct methods, exactly as it does for Cursor." />
+            best-for="Per-token AGGREGATOR that resells other makers' models behind one OpenAI-format endpoint and one key (openrouter.ai) — like Cursor's pool but paid per call, not a subscription. Reaches most of the catalog: supports-models is synced daily from openrouter.ai/models (OpenRouter's public model list): each catalogued model it lists as a text chat route, matched by name (absent today: composer-2.5, which is Cursor-only, and gemini-3-pro, which OpenRouter serves only as an image model). The operator should trust it LESS than a provider-direct method — prefer the maker's own method when the operator holds that key. BILLING: OpenRouter charges the maker's list price plus its platform fee (~5% on credit purchases at authoring), so the catalog price is a FLOOR for this method, never an exact estimate. ROUTING: a request may be served by any of OpenRouter's upstream hosts for that model; the MODEL-level jurisdiction filter (Step 0b) still governs which models are eligible, and this method's `us` code is OpenRouter's own (the operator's counterparty), NOT the upstream host's. Exposes the maker's reasoning dial where the routed model documents one (`reasoning_effort` on OpenAI / gpt-oss / GLM / DeepSeek families). Never the maker of any model it resells: the cross-provider BACKUP guard resolves makers through provider-direct methods, exactly as it does for Cursor." />
   </access-methods>
 
   <selection-algorithm>

{roadmodel-0.2.64 → roadmodel-0.2.67}/hatch_build.py

@@ -80,7 +80,15 @@ class BundleDocsHook(BuildHookInterface):  # type: ignore[misc]
         # source file: scripts/update_projects.py, which is also the bootstrap
         # URL, so it is mapped in rather than copied under src/.
         if self.target_name == "wheel":
-            updater = root / "scripts" / "update_projects.py"
-            if not updater.is_file():
-                raise FileNotFoundError(f"hatch_build.py: {updater} is missing")
-            force_include[str(updater)] = "roadmodel/update_projects.py"
+            for name in (
+                "update_projects",
+                "agent_context",
+                "agent_rules",
+                "upgrade_fleet",
+                "worktree_memory",
+                "updater_lock",
+            ):
+                updater = root / "scripts" / (name + ".py")
+                if not updater.is_file():
+                    raise FileNotFoundError(f"hatch_build.py: {updater} is missing")
+                force_include[str(updater)] = f"roadmodel/{name}.py"

{roadmodel-0.2.64 → roadmodel-0.2.67}/infra/README.md

@@ -153,7 +153,7 @@ is wired across three env scopes:
   Google Password Manager as `roadmodel ROADMODEL_IP_SALT`).
   Rotation cadence is **quarterly** per the Phase 7
   secrets-rotation policy. Defaults to a placeholder in
-  `web/lib/withRateLimit.ts` so local + CI builds work without
+  `web/lib/ip-salt.ts` so local + CI builds work without
   override, but production builds without the override silently
   bucket every IP under the same key, so the var **must** be
   set in every Vercel scope serving real traffic.
@@ -442,7 +442,9 @@ deliberate — Step 6 should not relitigate the Upstash decision.
 | `UPSTASH_REDIS_TOKEN`          | `roadmodel-web` Vercel env vars (Marketplace Upstash + alias) | Next.js rate limiter (live since Step 7 close-out)                          | web: production + preview + development + staging Custom Env (Step 7) | web: production + preview + development + staging |
 | `ROADMODEL_IP_SALT`            | `roadmodel-web` Vercel env vars (generated `openssl rand -hex 32`) | Next.js rate limiter daily IP+UA hashing salt (live since Step 7 close-out); rotate quarterly | web: production + preview + development + staging Custom Env (Step 7) | web: production + preview + development + staging |
 | `FRONTIER_ROADMAP_ENABLED`     | `roadmodel-web` Vercel env vars (default `false`)            | Next.js model-routing resolver — gates the Phase 5 Anthropic frontier branch | **not set anywhere in Phase 4 (defaults false)**         | web: production + preview + staging (set true on the scope where Phase 5 rolls out) |
-| `RECOMMEND_RATELIMIT_EXEMPT_USER_IDS` | `roadmodel-web` Vercel env vars (comma-separated founder Supabase uid; default empty) | Next.js rate limiter — user_ids that skip the `/api/recommend` daily limit (founder browser dogfooding); empty = nobody exempt | web: production (founder uid) | web: production (+ preview/staging if founder tests there) |
+| `RECOMMEND_RATELIMIT_EXEMPT_USER_IDS` | `roadmodel-web` Vercel env vars (comma-separated founder Supabase uid; default empty) | The FOUNDER list (`web/lib/funding-lane.ts`): these user_ids run `/api/recommend` and `/api/roadmap` on the operator's keys with no rate limit, and see the founder-only engines; empty = nobody | web: production (founder uid) | web: production (+ preview/staging if founder tests there) |
+| `RECOMMEND_INVITED_USER_IDS` | `roadmodel-web` Vercel env vars (comma-separated Supabase uids; default empty; personal data, never committed) | The INVITE list (`web/lib/funding-lane.ts`, Phase 4.11): these user_ids run on the operator's keys, 20 a day each; every caller on neither list gets `402 funding_required` before any paid call | web: production (Phase 4.11 Step 1) | web: production |
+| `ROADMODEL_DAILY_COST_CAP_USD` | `roadmodel-web` Vercel env vars (readable, not sensitive) | The operator lane's daily spend circuit breaker (`web/lib/spend-guard.ts`): once the UTC day's operator-funded `audit_log.cost_usd` reaches it, the operator lane answers 503 until midnight; 0 = off | web: production = `2` (Phase 4.11 Step 1) | web: production |
 
 Rules:
 

{roadmodel-0.2.64 → roadmodel-0.2.67}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "roadmodel"
-version = "0.2.64"
+version = "0.2.67"
 description = "BYO-key CLI that recommends the right AI model, platform, and settings for a prompt."
 readme = "README.md"
 requires-python = ">=3.11"
@@ -92,6 +92,11 @@ include = [
     "docs/catalog.json",
     "docs/benchmarks.json",
     "scripts/update_projects.py",
+    "scripts/agent_context.py",
+    "scripts/agent_rules.py",
+    "scripts/upgrade_fleet.py",
+    "scripts/worktree_memory.py",
+    "scripts/updater_lock.py",
     "hatch_build.py",
     "README.md",
     "LICENSE",