python-fastllm 0.0.48__tar.gz → 0.0.49__tar.gz

{python_fastllm-0.0.48 → python_fastllm-0.0.49}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-fastllm
-Version: 0.0.48
+Version: 0.0.49
 Author-email: Kerem Turgutlu <keremturgutlu@gmail.com>
 License: Apache-2.0
 Project-URL: Repository, https://github.com/AnswerDotAI/fastllm

{python_fastllm-0.0.48 → python_fastllm-0.0.49}/fastllm/__init__.py

@@ -2,4 +2,4 @@
 
 - `fastllm.responses`: Provider-independent Responses resources and streaming over FastLLM"""
 
-__version__ = "0.0.48"
+__version__ = "0.0.49"

{python_fastllm-0.0.48 → python_fastllm-0.0.49}/fastllm/_modidx.py

@@ -338,7 +338,7 @@ d = { 'settings': { 'branch': 'main',
                                'fastllm.types.Usage.__add__': ('types.html#usage.__add__', 'fastllm/types.py'),
                                'fastllm.types.Usage.__eq__': ('types.html#usage.__eq__', 'fastllm/types.py'),
                                'fastllm.types.Usage.__init__': ('types.html#usage.__init__', 'fastllm/types.py'),
-                               'fastllm.types._rm_ctx_tiers': ('types.html#_rm_ctx_tiers', 'fastllm/types.py'),
+                               'fastllm.types._flat_rates': ('types.html#_flat_rates', 'fastllm/types.py'),
                                'fastllm.types.approx_pricing': ('types.html#approx_pricing', 'fastllm/types.py'),
                                'fastllm.types.effort_code': ('types.html#effort_code', 'fastllm/types.py'),
                                'fastllm.types.effort_levels': ('types.html#effort_levels', 'fastllm/types.py'),

{python_fastllm-0.0.48 → python_fastllm-0.0.49}/fastllm/acomplete.py

@@ -67,10 +67,10 @@ def _codex_mtime_policy():
 
 # %% ../nbs/06_acomplete.ipynb #79075d95
 @flexicache(_codex_mtime_policy())
-def mk_client(model=None, vendor_name=None, api_name=None, api_key=None, base_url=None, xtra_hdrs=None,
+def mk_client(model=None, vendor_name=None, api_name=None, api_key=None, oauth_token=None, base_url=None, xtra_hdrs=None,
     timeout=httpx2.Timeout(connect=30, read=300, write=30, pool=10)):
     err_msg = f"please pass a valid one vendor: {', '.join(list(vendor_mapping))} or pass `api_name`,`base_url` and `api_key`"
-    if not vendor_name and not api_name and not (base_url and api_key):
+    if not vendor_name and not api_name and not (base_url and (api_key or oauth_token)):
         v, m = split_vendor(model)
         if v in vendor_mapping: vendor_name, model = v, m
     if vendor_name:
@@ -78,17 +78,17 @@ def mk_client(model=None, vendor_name=None, api_name=None, api_key=None, base_ur
         try:
             api_name, base_url, env_api_nm, *auth_json = vendor_mapping[vendor_name]
             base_url = override_base_url or base_url
-            if auth_json and not api_key and not os.getenv(env_api_nm):
+            if auth_json and not api_key and not oauth_token and not os.getenv(env_api_nm):
                 fn,keys = auth_json[0]  # pyright: ignore[reportAssignmentType]
                 auth_fn = Path(fn).expanduser()
                 if auth_fn.exists(): api_key = nested_idx(json.loads(auth_fn.read_text()), *keys)
-            api_key = get_api_key(api_key, env_api_nm)
+            if not oauth_token: api_key = get_api_key(api_key, env_api_nm)
         except KeyError: raise ValueError(f"Unknown vendor '{vendor_name}', {err_msg}")
-    elif base_url and api_key: vendor_name, api_name = ifnone(vendor_name, 'custom'), ifnone(api_name, 'openai_chat')
+    elif base_url and (api_key or oauth_token): vendor_name, api_name = ifnone(vendor_name, 'custom'), ifnone(api_name, 'openai_chat')
     elif (api_name:=infer_api_name(model)):  base_url, vendor_name = vendor_mapping[api_name][1], api_name
     else: raise ValueError(f"Model {model} can't be auto resolved, {err_msg}")
     api = api_registry[api_name]
-    hdrs = merge(api.get_hdrs(api_key), ifnone(xtra_hdrs, {}))
+    hdrs = merge(api.get_hdrs(api_key, oauth_token), ifnone(xtra_hdrs, {}))
     cli = AsyncHttpCli(base_url, base_headers=hdrs, timeout=timeout)
     return cli, api_name, vendor_name
 
@@ -155,18 +155,19 @@ async def _retry_stream(mk_gen, retries=2, retry_delay=0.5):
 
 # %% ../nbs/06_acomplete.ipynb #2379ec94
 @delegates(payload_kwargs)
-async def acomplete(msgs, model, api_name=None, vendor_name=None, api_key=None, base_url=None, endpoint=None, xtra_body=None, xtra_hdrs=None,
+async def acomplete(msgs, model, api_name=None, vendor_name=None, api_key=None, oauth_token=None, base_url=None, endpoint=None, xtra_body=None, xtra_hdrs=None,
     stream=False, previous_response_id=None, stop_callables=None, retries=2, retry_delay=0.5, **kwargs):
     "Unified completion across different APIs."
-    if not vendor_name and not api_name and not (base_url and api_key):
+    if api_key and oauth_token: raise ValueError('Pass `api_key` or `oauth_token`, not both: a request has one identity')
+    if not vendor_name and not api_name and not (base_url and (api_key or oauth_token)):
         v, m = split_vendor(model)
         if v in vendor_mapping: vendor_name, model = v, m
         elif v: api_name, model = v, m  # a registered transport api (e.g. claude_code): not an HTTP vendor
     own = api_name in api_registry and not hasattr(api_registry[api_name], 'endpoint') if api_name else False
     if own:
         vendor_name = ifnone(vendor_name, api_name)  # an own-transport api: no HTTP client to build
-        if api_key: kwargs['api_key'] = api_key  # its `mk_payload` applies a per-user credential itself
-    else: cli, api_name, vendor_name = mk_client(model=model, vendor_name=vendor_name, api_name=api_name, api_key=api_key, base_url=base_url, xtra_hdrs=xtra_hdrs)
+        if oauth_token: kwargs['oauth_token'] = oauth_token  # its `mk_payload` applies the user's token itself
+    else: cli, api_name, vendor_name = mk_client(model=model, vendor_name=vendor_name, api_name=api_name, api_key=api_key, oauth_token=oauth_token, base_url=base_url, xtra_hdrs=xtra_hdrs)
     api = api_registry[api_name]
     if previous_response_id is not None:
         if not getattr(api, 'supports_previous_response_id', False):

{python_fastllm-0.0.48 → python_fastllm-0.0.49}/fastllm/anthropic.py

@@ -322,8 +322,12 @@ def mk_payload(msgs, model, **kwargs):
     return payload
 
 # %% ../nbs/04_anthropic.ipynb #60d52c1f
-def get_hdrs(api_key=None):
-    return {"x-api-key": get_api_key(api_key, 'ANTHROPIC_API_KEY'), "anthropic-version": "2023-06-01"}
+def get_hdrs(api_key=None, oauth_token=None):
+    "An OAuth token travels as a bearer under Anthropic's OAuth beta; an API key as `x-api-key`"
+    hdrs = {"anthropic-version": "2023-06-01"}
+    if oauth_token: return hdrs | {"Authorization": f"Bearer {oauth_token}", "anthropic-beta": "oauth-2025-04-20"}
+    return hdrs | {"x-api-key": get_api_key(api_key, 'ANTHROPIC_API_KEY')}
+
 
 # %% ../nbs/04_anthropic.ipynb #0d03643a
 def cost(usage, m):

{python_fastllm-0.0.48 → python_fastllm-0.0.49}/fastllm/chat.py

@@ -168,6 +168,7 @@ class AsyncChat:
         api_name=None,            # API to use, one of ApiName: openai (responses), openai_chat, anthropic, gemini
         vendor_name=None,         # Vendor name, one of vendor_mapping which resolves api_base/api_key automatically
         api_key=None,             # API key when model can't be resolved or vendor_name is not known or codex
+        oauth_token=None,         # A subscription's OAuth token (Codex, Claude Code) in place of an API key
         base_url=None,            # API base url when model can't be resolved or vendor_name is not known
         endpoint=None,            # Override the transport's request path, for a server mounting Responses at a custom location
         extra_headers=None,       # Extra HTTP headers for custom providers
@@ -178,7 +179,7 @@ class AsyncChat:
         default_cbs=True          # Whether to include default callbacks
     ):
         "LiteLLM chat client."
-        if not vendor_name and not api_name and not (base_url and api_key):
+        if not vendor_name and not api_name and not (base_url and (api_key or oauth_token)):
             v, model = split_vendor(model)
             if v in vendor_mapping: vendor_name = v
             elif v: api_name = v  # a registered transport api (e.g. claude_code)
@@ -248,6 +249,7 @@ def _prep_call(self:AsyncChat, search, max_tokens, kwargs, stream=False, think=N
     if self.api_name:      kwargs['api_name'] = self.api_name
     if self.vendor_name:   kwargs['vendor_name'] = self.vendor_name
     if self.api_key:       kwargs['api_key'] = self.api_key
+    if self.oauth_token:   kwargs['oauth_token'] = self.oauth_token
     if self.base_url:      kwargs['base_url'] = self.base_url
     if self.endpoint:      kwargs['endpoint'] = self.endpoint
     if self.extra_headers: kwargs['xtra_hdrs'] = self.extra_headers

{python_fastllm-0.0.48 → python_fastllm-0.0.49}/fastllm/gemini.py

@@ -267,9 +267,12 @@ def mk_payload(msgs, model, **kwargs):
     return payload
 
 # %% ../nbs/05_gemini.ipynb #60d52c1f
-def get_hdrs(api_key=None):
+def get_hdrs(api_key=None, oauth_token=None):
+    "Bearer auth for an OAuth token; otherwise the API key header"
+    if oauth_token: return {"Authorization": f"Bearer {oauth_token}"}
     return {"x-goog-api-key": get_api_key(api_key, 'GEMINI_API_KEY')}
 
+
 # %% ../nbs/05_gemini.ipynb #4ee3891f
 def cost(usage, m):
     raw = usage.raw

{python_fastllm-0.0.48 → python_fastllm-0.0.49}/fastllm/openai_chat.py

@@ -198,7 +198,10 @@ def mk_payload(msgs, model, **kwargs):
     return payload
 
 # %% ../nbs/03_oai_chat.ipynb #16e813d2
-def get_hdrs(api_key=None): return {"Authorization": f"Bearer {get_api_key(api_key, 'OPENAI_API_KEY')}"}
+def get_hdrs(api_key=None, oauth_token=None):
+    "Bearer auth: an OAuth access token, or else an API key from the argument or the environment"
+    return {"Authorization": f"Bearer {oauth_token or get_api_key(api_key, 'OPENAI_API_KEY')}"}
+
 
 # %% ../nbs/03_oai_chat.ipynb #f89e2bf6
 def cost(usage, m):

{python_fastllm-0.0.48 → python_fastllm-0.0.49}/fastllm/openai_responses.py

@@ -256,7 +256,10 @@ def mk_payload(msgs, model, **kwargs):
     return payload
 
 # %% ../nbs/02_oai_responses.ipynb #529fd4bc
-def get_hdrs(api_key=None): return {"Authorization": f"Bearer {get_api_key(api_key, 'OPENAI_API_KEY')}"}
+def get_hdrs(api_key=None, oauth_token=None):
+    "Bearer auth: an OAuth access token, or else an API key from the argument or the environment"
+    return {"Authorization": f"Bearer {oauth_token or get_api_key(api_key, 'OPENAI_API_KEY')}"}
+
 
 # %% ../nbs/02_oai_responses.ipynb #a907ffa8
 def cost(usage, m):

{python_fastllm-0.0.48 → python_fastllm-0.0.49}/fastllm/types.py

@@ -300,14 +300,14 @@ register_model_info('muse-spark-1.1', vendor_name='meta_ai', **modern_llm, max_i
 codex_pricing = dict(input_cost_per_token = 0.10/1_000_000, output_cost_per_token = 0.50/1_000_000,
     cache_creation_input_token_cost = 0.10/1_000_000, cache_read_input_token_cost = 0.10/1_000_000)
 
-def _rm_ctx_tiers(vendor_name, model):
-    "Drop upstream context-tier rates, which would otherwise override flat subscription pricing"
+def _flat_rates(vendor_name, model):
+    "Keep only the flat subscription rates: drop the upstream context, cache-TTL and service-tier variants that would otherwise apply"
     info = model_info_registry[vendor_name, model]
-    for k in [k for k in info if re.search(r'_above_\d+k_tokens', k)]: del info[k]
+    for k in [k for k in info if 'cost' in k and k not in codex_pricing and k != 'search_context_cost_per_query']: del info[k]
 
 for model in (codex54, codex54m, codex55):
     register_model_info(model, 'codex', base=model, base_vendor_name='chatgpt', supports_web_search=True, max_input_tokens=256000, **codex_pricing)
-    _rm_ctx_tiers('codex', model)
+    _flat_rates('codex', model)
 
 register_model_info(codex53spark, 'codex', **codex_pricing,
     supports_vision=False, supports_image_input=False, supports_web_search=True, supports_reasoning=True, supports_function_calling=True,
@@ -318,6 +318,7 @@ register_model_info(codex53spark, 'codex', **codex_pricing,
 register_model_info(fable51, 'anthropic', base=fable5)
 for model in (haik45, sonn45, sonn46, sonn5, opus46, opus48, opus5, fable5, fable51):
     register_model_info(model, 'claude_code', base=model, base_vendor_name='anthropic', **codex_pricing)
+    _flat_rates('claude_code', model)
 
 # %% ../nbs/00_types.ipynb #bb0c4c2a
 sol,terra,luna = gpt56s = 'gpt-5.6-sol gpt-5.6-terra gpt-5.6-luna'.split()
@@ -325,7 +326,7 @@ sol,terra,luna = gpt56s = 'gpt-5.6-sol gpt-5.6-terra gpt-5.6-luna'.split()
 # Its window is smaller than the API's: 371,331 input tokens is accepted and 371,981 is not, on all three.
 for model in gpt56s:
     register_model_info(model, 'codex', base=model, base_vendor_name='openai', max_input_tokens=371_000, **codex_pricing)
-    _rm_ctx_tiers('codex', model)
+    _flat_rates('codex', model)
 
 # %% ../nbs/00_types.ipynb #24cc47ec
 def get_model_pricing(mn, vendor_name, million=True):

{python_fastllm-0.0.48 → python_fastllm-0.0.49}/python_fastllm.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-fastllm
-Version: 0.0.48
+Version: 0.0.49
 Author-email: Kerem Turgutlu <keremturgutlu@gmail.com>
 License: Apache-2.0
 Project-URL: Repository, https://github.com/AnswerDotAI/fastllm