mirror of
https://github.com/MODSetter/SurfSense.git
synced 2026-05-04 05:12:38 +02:00
feat: unified credits and its cost calculations
This commit is contained in:
DESKTOP-RTLN3BA\$punk
parent
451a98936e
commit
ae9d36d77f
61 changed files with 5835 additions and 272 deletions
|
|
@ -138,7 +138,11 @@ def load_global_image_gen_configs():
|
|||
try:
|
||||
with open(global_config_file, encoding="utf-8") as f:
|
||||
data = yaml.safe_load(f)
|
||||
return data.get("global_image_generation_configs", [])
|
||||
configs = data.get("global_image_generation_configs", []) or []
|
||||
for cfg in configs:
|
||||
if isinstance(cfg, dict):
|
||||
cfg.setdefault("billing_tier", "free")
|
||||
return configs
|
||||
except Exception as e:
|
||||
print(f"Warning: Failed to load global image generation configs: {e}")
|
||||
return []
|
||||
|
|
@ -153,7 +157,11 @@ def load_global_vision_llm_configs():
|
|||
try:
|
||||
with open(global_config_file, encoding="utf-8") as f:
|
||||
data = yaml.safe_load(f)
|
||||
return data.get("global_vision_llm_configs", [])
|
||||
configs = data.get("global_vision_llm_configs", []) or []
|
||||
for cfg in configs:
|
||||
if isinstance(cfg, dict):
|
||||
cfg.setdefault("billing_tier", "free")
|
||||
return configs
|
||||
except Exception as e:
|
||||
print(f"Warning: Failed to load global vision LLM configs: {e}")
|
||||
return []
|
||||
|
|
@ -254,6 +262,15 @@ def load_openrouter_integration_settings() -> dict | None:
|
|||
"anonymous_enabled_free", settings["anonymous_enabled"]
|
||||
)
|
||||
|
||||
# Image generation + vision LLM emission are opt-in (issue L).
|
||||
# OpenRouter's catalogue contains hundreds of image / vision
|
||||
# capable models; auto-injecting all of them into every
|
||||
# deployment would explode the model selector and surprise
|
||||
# operators upgrading from prior versions. Default to False so
|
||||
# admins must explicitly turn them on.
|
||||
settings.setdefault("image_generation_enabled", False)
|
||||
settings.setdefault("vision_enabled", False)
|
||||
|
||||
return settings
|
||||
except Exception as e:
|
||||
print(f"Warning: Failed to load OpenRouter integration settings: {e}")
|
||||
|
|
@ -296,10 +313,60 @@ def initialize_openrouter_integration():
|
|||
)
|
||||
else:
|
||||
print("Info: OpenRouter integration enabled but no models fetched")
|
||||
|
||||
# Image generation + vision LLM emissions are opt-in (issue L).
|
||||
# Both reuse the catalogue already cached by ``service.initialize``
|
||||
# so we don't make additional network calls here.
|
||||
if settings.get("image_generation_enabled"):
|
||||
try:
|
||||
image_configs = service.get_image_generation_configs()
|
||||
if image_configs:
|
||||
config.GLOBAL_IMAGE_GEN_CONFIGS.extend(image_configs)
|
||||
print(
|
||||
f"Info: OpenRouter integration added {len(image_configs)} "
|
||||
f"image-generation models"
|
||||
)
|
||||
except Exception as e:
|
||||
print(f"Warning: Failed to inject OpenRouter image-gen configs: {e}")
|
||||
|
||||
if settings.get("vision_enabled"):
|
||||
try:
|
||||
vision_configs = service.get_vision_llm_configs()
|
||||
if vision_configs:
|
||||
config.GLOBAL_VISION_LLM_CONFIGS.extend(vision_configs)
|
||||
print(
|
||||
f"Info: OpenRouter integration added {len(vision_configs)} "
|
||||
f"vision LLM models"
|
||||
)
|
||||
except Exception as e:
|
||||
print(f"Warning: Failed to inject OpenRouter vision-LLM configs: {e}")
|
||||
except Exception as e:
|
||||
print(f"Warning: Failed to initialize OpenRouter integration: {e}")
|
||||
|
||||
|
||||
def initialize_pricing_registration():
|
||||
"""
|
||||
Teach LiteLLM the per-token cost of every deployment in
|
||||
``config.GLOBAL_LLM_CONFIGS`` (OpenRouter dynamic models pulled
|
||||
from the OpenRouter catalogue + any operator-declared YAML pricing).
|
||||
|
||||
Must run AFTER ``initialize_openrouter_integration()`` so the
|
||||
OpenRouter catalogue is populated and BEFORE the first LLM call so
|
||||
``response_cost`` is available in ``TokenTrackingCallback``.
|
||||
|
||||
Failures are logged but never raised — startup must not be blocked
|
||||
by a missing pricing entry; the worst-case is the model debits 0.
|
||||
"""
|
||||
try:
|
||||
from app.services.pricing_registration import (
|
||||
register_pricing_from_global_configs,
|
||||
)
|
||||
|
||||
register_pricing_from_global_configs()
|
||||
except Exception as e:
|
||||
print(f"Warning: Failed to register LiteLLM pricing: {e}")
|
||||
|
||||
|
||||
def initialize_llm_router():
|
||||
"""
|
||||
Initialize the LLM Router service for Auto mode.
|
||||
|
|
@ -444,14 +511,54 @@ class Config:
|
|||
os.getenv("STRIPE_RECONCILIATION_BATCH_SIZE", "100")
|
||||
)
|
||||
|
||||
# Premium token quota settings
|
||||
PREMIUM_TOKEN_LIMIT = int(os.getenv("PREMIUM_TOKEN_LIMIT", "3000000"))
|
||||
# Premium credit (micro-USD) quota settings.
|
||||
#
|
||||
# Storage unit is integer micro-USD (1_000_000 = $1.00). The legacy
|
||||
# ``PREMIUM_TOKEN_LIMIT`` and ``STRIPE_TOKENS_PER_UNIT`` env vars are
|
||||
# still honoured for one release as fall-back values — the prior
|
||||
# $1-per-1M-tokens Stripe price means every existing value maps 1:1
|
||||
# to micros, so operators upgrading without changing their .env still
|
||||
# get correct behaviour. A startup deprecation warning fires below if
|
||||
# they're set.
|
||||
PREMIUM_CREDIT_MICROS_LIMIT = int(
|
||||
os.getenv("PREMIUM_CREDIT_MICROS_LIMIT")
|
||||
or os.getenv("PREMIUM_TOKEN_LIMIT", "5000000")
|
||||
)
|
||||
STRIPE_PREMIUM_TOKEN_PRICE_ID = os.getenv("STRIPE_PREMIUM_TOKEN_PRICE_ID")
|
||||
STRIPE_TOKENS_PER_UNIT = int(os.getenv("STRIPE_TOKENS_PER_UNIT", "1000000"))
|
||||
STRIPE_CREDIT_MICROS_PER_UNIT = int(
|
||||
os.getenv("STRIPE_CREDIT_MICROS_PER_UNIT")
|
||||
or os.getenv("STRIPE_TOKENS_PER_UNIT", "1000000")
|
||||
)
|
||||
STRIPE_TOKEN_BUYING_ENABLED = (
|
||||
os.getenv("STRIPE_TOKEN_BUYING_ENABLED", "FALSE").upper() == "TRUE"
|
||||
)
|
||||
|
||||
# Safety ceiling on the per-call premium reservation. ``stream_new_chat``
|
||||
# estimates an upper-bound cost from ``litellm.get_model_info`` x the
|
||||
# config's ``quota_reserve_tokens`` and clamps the result to this value
|
||||
# so a misconfigured "$1000/M" model can't lock the user's whole balance
|
||||
# on one call. Default $1.00 covers realistic worst-cases (Opus + 4K
|
||||
# reserve_tokens ≈ $0.36) with headroom.
|
||||
QUOTA_MAX_RESERVE_MICROS = int(os.getenv("QUOTA_MAX_RESERVE_MICROS", "1000000"))
|
||||
|
||||
if os.getenv("PREMIUM_TOKEN_LIMIT") and not os.getenv(
|
||||
"PREMIUM_CREDIT_MICROS_LIMIT"
|
||||
):
|
||||
print(
|
||||
"Warning: PREMIUM_TOKEN_LIMIT is deprecated; rename to "
|
||||
"PREMIUM_CREDIT_MICROS_LIMIT (1:1 numerical mapping under the "
|
||||
"current Stripe price). The old key will be removed in a "
|
||||
"future release."
|
||||
)
|
||||
if os.getenv("STRIPE_TOKENS_PER_UNIT") and not os.getenv(
|
||||
"STRIPE_CREDIT_MICROS_PER_UNIT"
|
||||
):
|
||||
print(
|
||||
"Warning: STRIPE_TOKENS_PER_UNIT is deprecated; rename to "
|
||||
"STRIPE_CREDIT_MICROS_PER_UNIT (1:1 numerical mapping). "
|
||||
"The old key will be removed in a future release."
|
||||
)
|
||||
|
||||
# Anonymous / no-login mode settings
|
||||
NOLOGIN_MODE_ENABLED = os.getenv("NOLOGIN_MODE_ENABLED", "FALSE").upper() == "TRUE"
|
||||
ANON_TOKEN_LIMIT = int(os.getenv("ANON_TOKEN_LIMIT", "500000"))
|
||||
|
|
@ -464,6 +571,35 @@ class Config:
|
|||
# Default quota reserve tokens when not specified per-model
|
||||
QUOTA_MAX_RESERVE_PER_CALL = int(os.getenv("QUOTA_MAX_RESERVE_PER_CALL", "8000"))
|
||||
|
||||
# Per-image reservation (in micro-USD) used by ``billable_call`` for the
|
||||
# ``POST /image-generations`` endpoint when the global config does not
|
||||
# override it. $0.05 covers realistic worst-cases for current OpenAI /
|
||||
# OpenRouter image-gen pricing. Bypassed entirely for free configs.
|
||||
QUOTA_DEFAULT_IMAGE_RESERVE_MICROS = int(
|
||||
os.getenv("QUOTA_DEFAULT_IMAGE_RESERVE_MICROS", "50000")
|
||||
)
|
||||
|
||||
# Per-podcast reservation (in micro-USD). One agent LLM call generating
|
||||
# a transcript, typically 5k-20k completion tokens. $0.20 covers a long
|
||||
# premium-model run. Tune via env.
|
||||
QUOTA_DEFAULT_PODCAST_RESERVE_MICROS = int(
|
||||
os.getenv("QUOTA_DEFAULT_PODCAST_RESERVE_MICROS", "200000")
|
||||
)
|
||||
|
||||
# Per-video-presentation reservation (in micro-USD). Fan-out of N
|
||||
# slide-scene generations (up to ``VIDEO_PRESENTATION_MAX_SLIDES=30``)
|
||||
# plus refine retries; can produce many premium completions. $1.00
|
||||
# covers worst-case. Tune via env.
|
||||
#
|
||||
# NOTE: this equals the existing ``QUOTA_MAX_RESERVE_MICROS`` default of
|
||||
# 1_000_000. The override path in ``billable_call`` bypasses the
|
||||
# per-call clamp in ``estimate_call_reserve_micros``, so this is the
|
||||
# *actual* hold — raising it via env is fine but means a single video
|
||||
# task can lock $1+ of credit.
|
||||
QUOTA_DEFAULT_VIDEO_PRESENTATION_RESERVE_MICROS = int(
|
||||
os.getenv("QUOTA_DEFAULT_VIDEO_PRESENTATION_RESERVE_MICROS", "1000000")
|
||||
)
|
||||
|
||||
# Abuse prevention: concurrent stream cap and CAPTCHA
|
||||
ANON_MAX_CONCURRENT_STREAMS = int(os.getenv("ANON_MAX_CONCURRENT_STREAMS", "2"))
|
||||
ANON_CAPTCHA_REQUEST_THRESHOLD = int(
|
||||
|
|
|
|||
|
|
@ -19,6 +19,24 @@
|
|||
# Structure matches NewLLMConfig:
|
||||
# - Model configuration (provider, model_name, api_key, etc.)
|
||||
# - Prompt configuration (system_instructions, citations_enabled)
|
||||
#
|
||||
# COST-BASED PREMIUM CREDITS:
|
||||
# Each premium config bills the user's USD-credit balance based on the
|
||||
# actual provider cost reported by LiteLLM. For models LiteLLM already
|
||||
# knows (most OpenAI/Anthropic/etc. names) you don't need to do anything.
|
||||
# For custom Azure deployment names (e.g. an in-house "gpt-5.4" deployment)
|
||||
# or any model LiteLLM doesn't have in its built-in pricing table, declare
|
||||
# per-token costs inline so they bill correctly:
|
||||
#
|
||||
# litellm_params:
|
||||
# base_model: "my-custom-azure-deploy"
|
||||
# # USD per token; e.g. 0.000003 == $3.00 per million input tokens
|
||||
# input_cost_per_token: 0.000003
|
||||
# output_cost_per_token: 0.000015
|
||||
#
|
||||
# OpenRouter dynamic models pull pricing automatically from OpenRouter's
|
||||
# API — no inline declaration needed. Models without resolvable pricing
|
||||
# debit $0 from the user's balance and log a WARNING.
|
||||
|
||||
# Router Settings for Auto Mode
|
||||
# These settings control how the LiteLLM Router distributes requests across models
|
||||
|
|
@ -292,6 +310,17 @@ openrouter_integration:
|
|||
free_rpm: 20
|
||||
free_tpm: 100000
|
||||
|
||||
# Image generation + vision LLM emission are OPT-IN. OpenRouter's catalogue
|
||||
# contains hundreds of image- and vision-capable models; turning these on
|
||||
# injects them into the global Image-Generation / Vision-LLM model
|
||||
# selectors alongside any static configs. Tier (free/premium) is derived
|
||||
# per model the same way it is for chat (`:free` suffix or zero pricing).
|
||||
# When a user picks a premium image/vision model the call debits the
|
||||
# shared $5 USD-cost-based premium credit pool — so leaving these off
|
||||
# avoids surprise quota burn on existing deployments. Default: false.
|
||||
image_generation_enabled: false
|
||||
vision_enabled: false
|
||||
|
||||
litellm_params:
|
||||
max_tokens: 16384
|
||||
system_instructions: ""
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue