crabcode 0.1.7__py3-none-any.whl → 0.1.8__py3-none-any.whl

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.
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.5
2
2
  Name: crabcode
3
- Version: 0.1.7
3
+ Version: 0.1.8
4
4
  Summary: CrabCode - AI coding assistant in the terminal
5
5
  Requires-Python: >=3.10
6
6
  Requires-Dist: anthropic>=0.40.0
@@ -11,7 +11,7 @@ crabcode_cli/session_picker.py,sha256=Pett41MV4xcg6vDR4Rbq8C0bh40u-GVF_cjQldNQIo
11
11
  crabcode_cli/render/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
12
12
  crabcode_cli/render/messages.py,sha256=n9OsbLIAU3FyO9gwGMMJld7zq-71N0caz9ecmE2kIjo,568
13
13
  crabcode_core/__init__.py,sha256=ArpgEg5xC3iR4DRT0HBg7DwU3K-r1MnmDOHpIb1b0E4,142
14
- crabcode_core/_version.py,sha256=NTuzp5GKvC1kBra8mTYyHzhWrod9nFOh9Rj9CZMMAZQ,82
14
+ crabcode_core/_version.py,sha256=wnwhPnAP1aSzJ69EjcIqL2fSUO0KufQNNkLPy4cTEVo,82
15
15
  crabcode_core/_windows_process.py,sha256=PjRQqZfLSHdvexANQYHZDQNxyu2s1Yaxt1YNlawwNIU,2949
16
16
  crabcode_core/agent_manager.py,sha256=Wu7kHfA5_zSTHYLemqOD85deblcwDdtvnZVE4rmq-Vc,71475
17
17
  crabcode_core/command_line.py,sha256=MWbAF4QUDVdwGM8cKoHYnIgr8yOgqk3Y_0ndt9f-t6Y,544
@@ -32,17 +32,17 @@ crabcode_core/turn_details.py,sha256=Km_qHTGz2OhnrRFJP9kuO02AcX1q-uTmR010vIh4WBE
32
32
  crabcode_core/usage.py,sha256=vNydrtitdH3Q690k-3x738LhKiJsXGcMw-ruOEL0csU,16307
33
33
  crabcode_core/utf8_sanitize.py,sha256=RRTGgwA9jqK-HLR1WUSFwgG4xJn7ZukTDu7M6Oi3fHI,936
34
34
  crabcode_core/api/__init__.py,sha256=IXs4hLgC4cT85TLdmS4KgtPKYDytadkwvrDyBshr5ag,674
35
- crabcode_core/api/anthropic_adapter.py,sha256=VrQPYxAkd1hyq0GzOC-YQYp7GWJC43DDOAAXEmnoYZw,26432
36
- crabcode_core/api/azure_adapter.py,sha256=pzh_4oq-Oqn8hmIU4orldtK9U_R_yONqc5BmoZHCiUE,2441
37
- crabcode_core/api/base.py,sha256=2fAiYazrzbk5kb4oKARClnePT4IVn9cF2tgxxoeKbP4,8579
38
- crabcode_core/api/codex_adapter.py,sha256=pgX8rzfdotUMQNDHWPiSW9QbmFrv0-MqPJ130ayCfNc,54775
35
+ crabcode_core/api/anthropic_adapter.py,sha256=YusxU59dh_wm87xVTSv09T-Cx-Azz_wUK0zgjGs2VqI,26486
36
+ crabcode_core/api/azure_adapter.py,sha256=2wFAMlcrpTqySD9A5QxgR0Klig3MthJI_SUgBCmo2L8,2452
37
+ crabcode_core/api/base.py,sha256=lwVcfz6byMu4Yw0w-Yi5dxHlO2PHXPaiA5OlnvySmhk,8538
38
+ crabcode_core/api/codex_adapter.py,sha256=NKQuR-C-6HB96-4ixPCdrB4tsT1ehNYLqMJ5lzhkb2E,54813
39
39
  crabcode_core/api/error_logging.py,sha256=2GMD8naR-Tk9Eki_i2GJQShAWGVvt6XZATs8SLXRR8Q,5326
40
40
  crabcode_core/api/gemini_adapter.py,sha256=RVk3zlLPqDtzG8w5psu4aSoFyFLua5qQx3subqz-5DQ,1130
41
- crabcode_core/api/model_discovery.py,sha256=f4s0dHSzpLXHpOxlBO-C14pSqwucSHmaCYDdBROYMTw,7194
41
+ crabcode_core/api/model_discovery.py,sha256=Yow48Sa1ZAePIZOq1FfHJ8Yhn2iHAF8yzIJgZY_IkUA,10396
42
42
  crabcode_core/api/model_info.py,sha256=hAP8bOiELV80E_sbZxXfOXriiBjpZ8nmTcYAMrp0x_0,8661
43
- crabcode_core/api/network.py,sha256=108VR8Ix67bklhZoV2FIPNtQ_jrI7JXUjS9A98p81-I,2438
43
+ crabcode_core/api/network.py,sha256=PziTbJI8D6EvBvN4OVHqNwGLbdkRBIFq1oXLfcZM-lc,2798
44
44
  crabcode_core/api/ollama_adapter.py,sha256=ZpTg11zOUGFE0J88qBXL7AvxDQU5PjWg2wWvm4uxUvY,1158
45
- crabcode_core/api/openai_adapter.py,sha256=HRrsmoc01qEo49hXU2VBbEZiMgUPvwXxgBtx3s_0NHo,12303
45
+ crabcode_core/api/openai_adapter.py,sha256=Ea7W543mtgutmdKC7H5GbSIZg1X4bWvur3Ul759Zg_Y,12364
46
46
  crabcode_core/api/registry.py,sha256=tCVRE74bMUn9ukCbGp3bDtSOYZ6_EmsNfdVnnkGN6k4,2413
47
47
  crabcode_core/compact/__init__.py,sha256=9445RwiG5BI8b6mrXMcNkcPLs5NH3RSKFxDrXsgmmK8,395
48
48
  crabcode_core/compact/compact.py,sha256=_5BjLhPJA7WkvQzBAor7d9QXF9Nluui7bZTtBNvPLpY,19580
@@ -158,7 +158,7 @@ crabcode_gateway/event_bus.py,sha256=lWLCeoKkJ9fhDT_Kwc_esokDsbeRkPi5hsK87kPxRRY
158
158
  crabcode_gateway/middleware.py,sha256=bn0pUJHWINc_28ZgZ6e7i-AarkqhfQZMaxbknAVU5UA,11293
159
159
  crabcode_gateway/protocol.py,sha256=psXRNQjkfwliXZdETlo1qsGM0IcE_OIUarCLnSNuq6o,466
160
160
  crabcode_gateway/safe_file.py,sha256=uOdkxWdZNQ5RJ2jJ9urnIRiF_2uAP8Dkj8wR08vbDaU,2108
161
- crabcode_gateway/schemas.py,sha256=v4D5J6_tcsn9whiwpAYwg5a-kkQlF1j9YIwQiSZ0EvQ,56362
161
+ crabcode_gateway/schemas.py,sha256=iNmsQhYV1nZkblV6IR2qzZ-q0Bl9NMFf8SlmjCT1hEM,57381
162
162
  crabcode_gateway/server.py,sha256=ShCrO1Zw-Ed37HfUYihyOToJasGJjL7Kzy1BFA4CJ64,20408
163
163
  crabcode_gateway/session_registry.py,sha256=HhvLvlGDx0EgmGwjk2zHCCtWURmNDZHHmpBjroX35fk,1214
164
164
  crabcode_gateway/task_registry.py,sha256=Ij8USxQ13ockqhoZ286hn8jXckFeLglNSgwgcnQ5tX4,26461
@@ -177,7 +177,7 @@ crabcode_gateway/routes/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG
177
177
  crabcode_gateway/routes/agent.py,sha256=ZI97BVRMPgwpv45IqWFoESO_Khr0uFfy6J-tDojNWOU,13955
178
178
  crabcode_gateway/routes/auth.py,sha256=kIynPq7LUBlKwS--1Qp1hJHpLqebB5Unt1qAI8V_dUI,4945
179
179
  crabcode_gateway/routes/computer_use.py,sha256=_NtMi3zciudhDRcha53fXL3jECp6Oj5cWWvWKD_JLAY,4622
180
- crabcode_gateway/routes/config.py,sha256=VCA97t7dU8c5EctjHFsXotiFAE0vZ8N5K0ie97cQxZ4,70636
180
+ crabcode_gateway/routes/config.py,sha256=rcfM2ZKVd1tVET_x6oOLyNwZ6Y4gWl6iqnWXn2WWOxM,71251
181
181
  crabcode_gateway/routes/document.py,sha256=j01jt88A29GF6fHYcX2AparZAOPsUrwllec9YGtFswg,78665
182
182
  crabcode_gateway/routes/event.py,sha256=WOr1WHJ-5nCcpCWIkpIos2aAl2ASJUyYKSaLZ5bCRS0,155226
183
183
  crabcode_gateway/routes/git_review.py,sha256=-fbMOWo8JKnHCMsKew3i02lTE7Ohj9w6ocgFKCsYeio,10460
@@ -198,7 +198,7 @@ crabcode_search/embedder.py,sha256=NlNuOHS7t4JvaWXkNU-T_xxnkjRxFrr3l57dlB5IRkc,1
198
198
  crabcode_search/indexer.py,sha256=bBL3OE6HMqdIRBCnI--XjaHl7_G3H4VO2d-RMKtjlkI,7745
199
199
  crabcode_search/store.py,sha256=s1hsyRRGcAO_hdJli6M0C8rhkMW4tXj5ZVMPuCP3uhQ,8074
200
200
  crabcode_search/tool.py,sha256=xJjAN3esvJIH3z7ILNQ1gQs5Ty5fu6u7CYX6P2WB_BY,10238
201
- crabcode-0.1.7.dist-info/METADATA,sha256=Q6Sqy5M53yQ7A7I6vBjUJE-vVeXQTju-vninazwV9xs,2799
202
- crabcode-0.1.7.dist-info/WHEEL,sha256=W3fkpkm7-wf9vBI5Z-7s0eWkeM-spu78I8Neb98DeEg,87
203
- crabcode-0.1.7.dist-info/entry_points.txt,sha256=tpYKME-yDLFb2eqwryWJG-d-HW4dkvDbzwogRl2jays,52
204
- crabcode-0.1.7.dist-info/RECORD,,
201
+ crabcode-0.1.8.dist-info/METADATA,sha256=JrxGaP-8IR-KHrZ5XS-kCxVE0ks51VfgaDSOnSzHO0k,2799
202
+ crabcode-0.1.8.dist-info/WHEEL,sha256=W3fkpkm7-wf9vBI5Z-7s0eWkeM-spu78I8Neb98DeEg,87
203
+ crabcode-0.1.8.dist-info/entry_points.txt,sha256=tpYKME-yDLFb2eqwryWJG-d-HW4dkvDbzwogRl2jays,52
204
+ crabcode-0.1.8.dist-info/RECORD,,
crabcode_core/_version.py CHANGED
@@ -1,3 +1,3 @@
1
1
  """Single source of truth for the CrabCode release version."""
2
2
 
3
- VERSION = "0.1.7"
3
+ VERSION = "0.1.8"
@@ -123,6 +123,7 @@ def _tools_to_api(tools: list[dict[str, Any]]) -> list[dict[str, Any]]:
123
123
  class AnthropicAdapter(APIAdapter):
124
124
  """Adapter for Anthropic's Messages API (direct, first-party)."""
125
125
 
126
+ _sdk = anthropic
126
127
  emits_response_item_events = True
127
128
 
128
129
  def __init__(self, config: ApiConfig):
@@ -143,7 +144,7 @@ class AnthropicAdapter(APIAdapter):
143
144
  kwargs["default_headers"] = config.http_headers
144
145
 
145
146
  from crabcode_core.api.network import sdk_options
146
- self.client = anthropic.AsyncAnthropic(**kwargs, **sdk_options(config))
147
+ self.client = anthropic.AsyncAnthropic(**kwargs, **sdk_options(config, self._sdk))
147
148
  self._api_key = api_key
148
149
 
149
150
  def _messages_url(self) -> str | None:
@@ -632,7 +633,7 @@ class BedrockAdapter(AnthropicAdapter):
632
633
  def __init__(self, config: ApiConfig):
633
634
  self.config = config
634
635
  from crabcode_core.api.network import sdk_options
635
- self.client = anthropic.AsyncAnthropicBedrock(**sdk_options(config))
636
+ self.client = anthropic.AsyncAnthropicBedrock(**sdk_options(config, self._sdk))
636
637
 
637
638
 
638
639
  class VertexAdapter(AnthropicAdapter):
@@ -644,5 +645,5 @@ class VertexAdapter(AnthropicAdapter):
644
645
  self.client = anthropic.AsyncAnthropicVertex(
645
646
  region=os.environ.get("CLOUD_ML_REGION", "us-east5"),
646
647
  project_id=os.environ.get("ANTHROPIC_VERTEX_PROJECT_ID"),
647
- **sdk_options(config),
648
+ **sdk_options(config, self._sdk),
648
649
  )
@@ -57,7 +57,7 @@ class AzureOpenAIAdapter(OpenAIAdapter):
57
57
  kwargs["default_headers"] = config.http_headers
58
58
 
59
59
  from crabcode_core.api.network import sdk_options
60
- return openai.AsyncAzureOpenAI(**kwargs, **sdk_options(config))
60
+ return openai.AsyncAzureOpenAI(**kwargs, **sdk_options(config, self._sdk))
61
61
 
62
62
  def _resolve_model(self, config: ModelConfig) -> str:
63
63
  """Azure uses deployment names instead of model names."""
crabcode_core/api/base.py CHANGED
@@ -188,13 +188,13 @@ class APIAdapter(ABC):
188
188
  async def reset_network_client(self) -> None:
189
189
  """Replace a failed SDK connection pool without changing its routing policy."""
190
190
  import asyncio
191
- from crabcode_core.api.network import http_options, request_timeout
191
+ from crabcode_core.api.network import sdk_http_client
192
192
 
193
193
  client = getattr(self, "client", None)
194
194
  copy = getattr(client, "with_options", None)
195
195
  if not callable(copy):
196
196
  return
197
- transport = httpx.AsyncClient(timeout=request_timeout(self.config), **http_options(self.config))
197
+ transport = sdk_http_client(self.config, getattr(self, "_sdk", None))
198
198
  try:
199
199
  replacement = copy(http_client=transport, max_retries=0)
200
200
  except BaseException:
@@ -486,6 +486,7 @@ class CodexAdapter(APIAdapter):
486
486
  def __init__(self, config: ApiConfig):
487
487
  import openai
488
488
 
489
+ self._sdk = openai
489
490
  self.config = config
490
491
  api_key = None
491
492
  oauth_account_id = None
@@ -524,7 +525,7 @@ class CodexAdapter(APIAdapter):
524
525
  # differing built-in policy (which also retries 429) and apply the
525
526
  # configured policy explicitly in _create_sdk_response_stream().
526
527
  from crabcode_core.api.network import sdk_options
527
- kwargs.update(sdk_options(config))
528
+ kwargs.update(sdk_options(config, self._sdk))
528
529
  self.client = openai.AsyncOpenAI(**kwargs)
529
530
 
530
531
  async def resolve_context_window(self) -> int:
@@ -6,6 +6,7 @@ import asyncio
6
6
  import json
7
7
  import os
8
8
  from collections.abc import Mapping
9
+ from typing import Any
9
10
  from urllib.parse import urlsplit, urlunsplit
10
11
 
11
12
  import httpx
@@ -18,6 +19,63 @@ class ModelDiscoveryError(ValueError):
18
19
  """A safe, user-facing discovery error (never a raw provider response)."""
19
20
 
20
21
 
22
+ def _token_limit(*values: Any) -> int | None:
23
+ # Metadata is optional: ignore malformed values without losing the catalog.
24
+ limits = [value for value in values if type(value) is int and 0 < value <= 2_147_483_647]
25
+ return min(limits) if limits else None
26
+
27
+
28
+ def _model_metadata(entry: Mapping[str, Any]) -> dict[str, Any]:
29
+ """Normalize explicit limits; never guess capabilities from a model name."""
30
+ # https://platform.claude.com/docs/en/api/models/list
31
+ # https://openrouter.ai/docs/api/api-reference/models/list-all-models-and-their-properties
32
+ top = entry.get("top_provider")
33
+ top = top if isinstance(top, dict) else {}
34
+ result: dict[str, Any] = {}
35
+ context = _token_limit(*(entry.get(key) for key in (
36
+ "context_length", "context_window", "max_context_length", "max_model_len", "max_input_tokens",
37
+ )), top.get("context_length"))
38
+ output = _token_limit(*(entry.get(key) for key in (
39
+ "max_output_tokens", "max_completion_tokens", "max_tokens",
40
+ )), top.get("max_completion_tokens"))
41
+ if context is not None:
42
+ result["context_window"] = context
43
+ if output is not None:
44
+ result["max_tokens"] = output
45
+ capabilities = entry.get("capabilities")
46
+ thinking = capabilities.get("thinking") if isinstance(capabilities, dict) else None
47
+ supported = thinking.get("supported") if isinstance(thinking, dict) else None
48
+ if type(supported) is bool:
49
+ result["thinking_enabled"] = supported
50
+ return result
51
+
52
+
53
+ def model_import_config(metadata: Mapping[str, Any], group: ApiConfig) -> dict[str, Any]:
54
+ """Apply only supported settings, retaining explicit group preferences."""
55
+ result: dict[str, Any] = {}
56
+ context = _token_limit(metadata.get("context_window"))
57
+ if context is not None and group.context_window is None:
58
+ result["context_window"] = context
59
+ output = _token_limit(metadata.get("max_tokens"))
60
+ if output is not None:
61
+ # A deliberately smaller output budget stays in force. Otherwise use
62
+ # the published model limit instead of the generic 16k default.
63
+ result["max_tokens"] = min(output, group.max_tokens) if "max_tokens" in group.model_fields_set else output
64
+ thinking = metadata.get("thinking_enabled")
65
+ if type(thinking) is bool:
66
+ result["thinking_enabled"] = thinking and group.thinking_enabled
67
+ protocol = (group.format or "openai") if group.provider == "router" else (group.provider or "anthropic")
68
+ if protocol == "anthropic" and output is not None and result.get("thinking_enabled", group.thinking_enabled):
69
+ # Anthropic budgeted thinking requires at least 1024 tokens, strictly
70
+ # below max_tokens. Importing a smaller limit must not break requests.
71
+ budget = min(group.thinking_budget, result["max_tokens"] - 1)
72
+ if budget < 1024:
73
+ result["thinking_enabled"] = False
74
+ elif budget != group.thinking_budget:
75
+ result["thinking_budget"] = budget
76
+ return result
77
+
78
+
21
79
  def _connection(config: ApiConfig, env: Mapping[str, str]) -> tuple[str, httpx.Headers, bool]:
22
80
  provider = config.provider or "anthropic"
23
81
  protocol = (config.format or "openai") if provider == "router" else provider
@@ -63,9 +121,9 @@ def _connection(config: ApiConfig, env: Mapping[str, str]) -> tuple[str, httpx.H
63
121
  return endpoint, headers, anthropic
64
122
 
65
123
 
66
- async def _fetch(config: ApiConfig, env: Mapping[str, str]) -> list[dict[str, str]]:
124
+ async def _fetch(config: ApiConfig, env: Mapping[str, str]) -> list[dict[str, Any]]:
67
125
  endpoint, headers, anthropic = _connection(config, env)
68
- models: dict[str, dict[str, str]] = {}
126
+ models: dict[str, dict[str, Any]] = {}
69
127
  cursors: set[str] = set()
70
128
  params: dict[str, str | int] = {"limit": 1000} if anthropic else {}
71
129
  remaining_bytes = 4 * 1024 * 1024
@@ -93,10 +151,13 @@ async def _fetch(config: ApiConfig, env: Mapping[str, str]) -> list[dict[str, st
93
151
  if not isinstance(model_id, str) or not model_id.strip() or len(model_id) > 512 or any(ord(char) < 32 for char in model_id):
94
152
  raise ModelDiscoveryError("模型列表包含无效的模型 ID,请检查服务端响应")
95
153
  model_id = model_id.strip()
96
- model = {"id": model_id}
97
- display_name = entry.get("display_name")
154
+ model: dict[str, Any] = {"id": model_id}
155
+ display_name = entry.get("display_name") or entry.get("name")
98
156
  if isinstance(display_name, str) and display_name.strip():
99
157
  model["name"] = display_name.strip()[:512]
158
+ settings = model_import_config(_model_metadata(entry), config)
159
+ if settings:
160
+ model["config"] = settings
100
161
  models.setdefault(model_id, model)
101
162
  if len(models) > 5000:
102
163
  raise ModelDiscoveryError("模型列表过大,请缩小服务端返回范围后重试")
@@ -110,7 +171,7 @@ async def _fetch(config: ApiConfig, env: Mapping[str, str]) -> list[dict[str, st
110
171
  raise ModelDiscoveryError("模型列表分页过多,请缩小服务端返回范围后重试")
111
172
 
112
173
 
113
- async def discover_models(config: ApiConfig, settings_env: Mapping[str, str] | None = None) -> list[dict[str, str]]:
174
+ async def discover_models(config: ApiConfig, settings_env: Mapping[str, str] | None = None) -> list[dict[str, Any]]:
114
175
  try:
115
176
  # Match session initialization's setdefault precedence without mutating
116
177
  # the Gateway environment or requiring an initialized chat session.
@@ -26,17 +26,24 @@ def http_options(config: Any) -> dict[str, Any]:
26
26
  return {"trust_env": False, "proxy": proxy}
27
27
 
28
28
 
29
- def request_timeout(config: Any) -> httpx.Timeout:
29
+ def request_timeout(config: Any, sdk: Any = None) -> Any:
30
30
  timeout = float(getattr(config, "timeout", 300))
31
- return httpx.Timeout(timeout, connect=min(timeout, 10.0))
31
+ # SDK exports follow their supported transport (httpx or httpx2).
32
+ timeout_type = sdk.Timeout if sdk is not None else httpx.Timeout
33
+ return timeout_type(timeout, connect=min(timeout, 10.0))
32
34
 
33
35
 
34
- def sdk_options(config: Any) -> dict[str, Any]:
36
+ def sdk_http_client(config: Any, sdk: Any = None) -> Any:
37
+ client_type = sdk.DefaultAsyncHttpxClient if sdk is not None else httpx.AsyncClient
38
+ return client_type(**http_options(config), timeout=request_timeout(config, sdk))
39
+
40
+
41
+ def sdk_options(config: Any, sdk: Any) -> dict[str, Any]:
35
42
  # The query loop owns retry budgets; nested SDK retries hide progress.
36
- options: dict[str, Any] = {"max_retries": 0, "timeout": request_timeout(config)}
43
+ options: dict[str, Any] = {"max_retries": 0, "timeout": request_timeout(config, sdk)}
37
44
  routing = http_options(config)
38
45
  if routing:
39
- options["http_client"] = httpx.AsyncClient(**routing, timeout=request_timeout(config))
46
+ options["http_client"] = sdk_http_client(config, sdk)
40
47
  return options
41
48
 
42
49
 
@@ -168,6 +168,9 @@ class OpenAIAdapter(APIAdapter):
168
168
  """
169
169
 
170
170
  def __init__(self, config: ApiConfig):
171
+ import openai
172
+
173
+ self._sdk = openai
171
174
  self.config = config
172
175
  self.client = self._create_client(config)
173
176
 
@@ -190,7 +193,7 @@ class OpenAIAdapter(APIAdapter):
190
193
  kwargs["default_headers"] = config.http_headers
191
194
 
192
195
  from crabcode_core.api.network import sdk_options
193
- return openai.AsyncOpenAI(**kwargs, **sdk_options(config))
196
+ return openai.AsyncOpenAI(**kwargs, **sdk_options(config, self._sdk))
194
197
 
195
198
  def _resolve_model(self, config: ModelConfig) -> str:
196
199
  """Resolve the model name for the API call. Override in subclasses."""
@@ -29,6 +29,7 @@ from crabcode_gateway.schemas import (
29
29
  GoalRequest,
30
30
  GoalState,
31
31
  ModelInfo,
32
+ ModelDiscoveryResponse,
32
33
  ModelSettingsEntry,
33
34
  ModelSettingsMutationRequest,
34
35
  ModelSettingsResponse,
@@ -66,7 +67,7 @@ class GroupModelsRequest(BaseModel):
66
67
  cwd: str | None = None
67
68
 
68
69
 
69
- @router.post("/config/discover-models")
70
+ @router.post("/config/discover-models", response_model=ModelDiscoveryResponse, response_model_exclude_none=True)
70
71
  async def discover_group_models(req: GroupModelsRequest, request: Request) -> dict[str, Any]:
71
72
  from crabcode_core.api.model_discovery import ModelDiscoveryError, discover_models
72
73
 
@@ -316,7 +317,10 @@ def _redact_model_settings(value: Any, key: str = "") -> Any:
316
317
  """Keep useful configuration visible without echoing embedded credentials."""
317
318
  normalized_key = key.lower().replace("-", "_")
318
319
  is_reference = normalized_key.endswith("_env") or normalized_key.endswith("_path")
319
- if key and not is_reference and any(part in normalized_key for part in _SENSITIVE_CONFIG_KEYS):
320
+ is_token_count = type(value) is int and normalized_key in {
321
+ "max_tokens", "max_output_tokens", "max_completion_tokens", "budget_tokens", "compact_buffer_tokens",
322
+ }
323
+ if key and not is_reference and not is_token_count and any(part in normalized_key for part in _SENSITIVE_CONFIG_KEYS):
320
324
  return "[redacted]"
321
325
  if isinstance(value, dict):
322
326
  return {
@@ -626,11 +630,13 @@ def _mutate_model_settings(request: Request, req: ModelSettingsMutationRequest)
626
630
 
627
631
  if action == "import_group_models":
628
632
  import re
633
+ from crabcode_core.api.model_discovery import model_import_config
629
634
 
630
635
  name = _validate_model_settings_name(req.name)
631
636
  preview = _model_settings_from_files(cwd)
632
637
  if name not in preview.groups:
633
638
  raise HTTPException(status_code=404, detail="配置组不存在,请刷新模型目录")
639
+ group = ConfigManager(cwd=cwd).load().groups[name]
634
640
  models = current.setdefault("models", {})
635
641
  if not isinstance(models, dict):
636
642
  raise HTTPException(status_code=422, detail="models must be a JSON object")
@@ -654,7 +660,9 @@ def _mutate_model_settings(request: Request, req: ModelSettingsMutationRequest)
654
660
  tail = f" ({suffix})"
655
661
  candidate = stem[:120 - len(tail)] + tail
656
662
  suffix += 1
657
- models[candidate] = {"group": name, "model": model_id}
663
+ metadata = req.model_metadata.get(model_id)
664
+ settings = model_import_config(metadata.model_dump(exclude_none=True), group) if metadata else {}
665
+ models[candidate] = {"group": name, "model": model_id, **settings}
658
666
  used_names.add(candidate)
659
667
  elif action in {"upsert_model", "delete_model"}:
660
668
  name = _validate_model_settings_name(req.name)
@@ -277,6 +277,27 @@ class SetPermissionModeRequest(BaseModel):
277
277
  session_id: str | None = None
278
278
 
279
279
 
280
+ class DiscoveredModelSettings(BaseModel):
281
+ """Allowlisted settings from the model catalog, never connection credentials."""
282
+
283
+ model_config = {"extra": "forbid"}
284
+ context_window: int | None = Field(default=None, strict=True, gt=0, le=2_147_483_647)
285
+ max_tokens: int | None = Field(default=None, strict=True, gt=0, le=2_147_483_647)
286
+ thinking_enabled: bool | None = Field(default=None, strict=True)
287
+ thinking_budget: int | None = Field(default=None, strict=True, ge=1024, le=2_147_483_647)
288
+
289
+
290
+ class ModelDiscoveryEntry(BaseModel):
291
+ id: str
292
+ name: str | None = None
293
+ config: DiscoveredModelSettings | None = None
294
+
295
+
296
+ class ModelDiscoveryResponse(BaseModel):
297
+ group: str
298
+ models: list[ModelDiscoveryEntry]
299
+
300
+
280
301
  class ModelSettingsMutationRequest(BaseModel):
281
302
  """A focused mutation of the named model configuration catalog."""
282
303
 
@@ -296,6 +317,7 @@ class ModelSettingsMutationRequest(BaseModel):
296
317
  config: dict[str, Any] | None = None
297
318
  remove_fields: list[str] = Field(default_factory=list)
298
319
  model_ids: list[str] = Field(default_factory=list, max_length=1000)
320
+ model_metadata: dict[str, DiscoveredModelSettings] = Field(default_factory=dict, max_length=1000)
299
321
 
300
322
  @model_validator(mode="after")
301
323
  def validate_mutation(self) -> "ModelSettingsMutationRequest":
@@ -310,6 +332,8 @@ class ModelSettingsMutationRequest(BaseModel):
310
332
  for value in self.model_ids
311
333
  ):
312
334
  raise ValueError("Select valid model IDs to add to the group")
335
+ if not self.model_metadata.keys() <= {value.strip() for value in self.model_ids}:
336
+ raise ValueError("Model metadata must belong to the selected model IDs")
313
337
  if self.name is not None:
314
338
  self.name = self.name.strip()
315
339
  return self