github-featured/dayuanjiang--next-ai-draw-io
2
0
Fork 0
Code Issues Pull Requests Deployments Models Agent Notes Packages Projects Releases Wiki Activity
Files
main
dayuanjiang--next-ai-draw-io/env.example
T
Dayuan Jiang 155ef4f7ac fix: raise the output budget so reasoning models reach the tool call (#927) 
* fix: raise the output budget so reasoning models reach the tool call

A reasoning model spends the output budget in order: thinking first, then prose,
then the tool call. With 16000 the thinking alone can consume all of it, so the
turn ends with finishReason "length" before display_diagram is ever called. The
canvas stays empty and nothing surfaces in the UI, because no tool call means no
tool error, and the client never reads finishReason.

Measured on openrouter deepseek/deepseek-v4-flash, the model from the report:
- max_tokens=800 with reasoning on returns reasoning_tokens=800, empty content,
  finish_reason length. So reasoning is billed against this budget, not exempt.
- refining an existing diagram (19k chars of XML in the input) produced 49142
  chars of reasoning, zero tool calls, finishReason "length" at 16000
- the same request at 40000 finished and called edit_diagram with 12 operations

64000 cannot just be sent to every model: bedrock claude-3-haiku caps at 4096,
nova-lite at 10000, and the openrouter deepseek-r1 endpoint counts input and
output against one 64000 ceiling. All three name the real limit in the 400, so
parse it and retry once. Verified: nova-lite logs "64000 rejected, retrying with
10000" and then completes its tool call.

Also expose the budget in Settings. It is sent as a header rather than read from
env only, so desktop users can raise it themselves without an env file.

vercel.json goes back to the 300s it had before #238 traded it for $2-4/month.
That is now Vercel's own default, and billing pauses while the function waits on
the model, so the saving that motivated 120s no longer applies. edgeone.json is
left alone: its 120 may be that platform's actual ceiling.

* fix: only reinterpret an error as a budget rejection when it says so

Review of the first commit found the retry could fire on errors that have
nothing to do with the budget, which would replace a readable provider error
with a truncated response: exactly the symptom this PR exists to remove.

- Drop the generic "lower than N" pattern. For the Bedrock message it was dead
  code, since "model limit of N" matches first with the same number. Left live,
  it would read a number out of any message shaped like "must be lower than 2".
- Skip errors whose status is not 400 or 422, so auth and rate-limit failures
  are never reinterpreted.
- Require the parsed ceiling to be at least 1024. Below that a diagram cannot
  come out whole, so retrying would hide the error behind broken XML.
- Validate MAX_OUTPUT_TOKENS from env the same way as the header, so a stray
  "-1" falls back instead of reaching the provider.

Adds tests for the retry wrapper itself, which had none: it retries once with
the named ceiling, leaves a 401 alone, does not retry when the ceiling is not
smaller, propagates a second rejection, and preserves the other call options.

Re-verified against the live APIs: bedrock nova-lite still logs "64000 rejected,
retrying with 10000" and completes its tool call, and deepseek-v4-flash still
finishes normally at 64000.
2026-08-22 20:47:10 +09:00

207 lines
10 KiB
Plaintext
Raw Permalink Blame History

# AI Provider Configuration
# AI_PROVIDER: Which provider to use
# Options: bedrock, openai, anthropic, google, vertexai, azure, ollama, openrouter, aihubmix, deepseek, siliconflow, gateway, novita
# Default: bedrock
AI_PROVIDER=bedrock
# AI_MODEL: The model ID for your chosen provider (REQUIRED)
# Tip: For a single-provider quick multi-model setup, list comma-separated model IDs.
# The first one becomes the default and the rest appear in the model picker.
# For multiple providers or custom apiKeyEnv/baseUrlEnv, use AI_MODELS_CONFIG / ai-models.json instead.
# Example: AI_MODEL=doubao-seed-1-8-251215,doubao-seed-1-6-flash,doubao-seed-1-6-pro
AI_MODEL=global.anthropic.claude-sonnet-4-5-20250929-v1:0
# Output limit, all providers (default: 64000). Shared by reasoning and the diagram XML,
# so a thinking model can spend it all before the tool call. Users can override it in Settings.
# If a model's own ceiling is lower, the request is retried with that ceiling automatically.
# MAX_OUTPUT_TOKENS=64000
# AWS Bedrock Configuration
# AWS_REGION=us-east-1
# AWS_ACCESS_KEY_ID=your-access-key-id
# AWS_SECRET_ACCESS_KEY=your-secret-access-key
# Note: Claude and Nova models support reasoning/extended thinking
# BEDROCK_REASONING_BUDGET_TOKENS=12000 # Optional: Claude reasoning budget in tokens (1024-64000)
# BEDROCK_REASONING_EFFORT=medium # Optional: Nova reasoning effort (low/medium/high)
# OpenAI Configuration
# OPENAI_API_KEY=sk-...
# OPENAI_BASE_URL=https://api.openai.com/v1 # Optional: Custom OpenAI-compatible endpoint
# OPENAI_ORGANIZATION=org-... # Optional
# OPENAI_PROJECT=proj_... # Optional
# Note: o1/o3/gpt-5 models automatically enable reasoning summary (default: detailed)
# OPENAI_REASONING_EFFORT=low # Optional: Reasoning effort (minimal/low/medium/high) - for o1/o3/gpt-5
# OPENAI_REASONING_SUMMARY=detailed # Optional: Override reasoning summary (none/brief/detailed)
# Anthropic (Direct) Configuration
# ANTHROPIC_API_KEY=sk-ant-... # Sent as `x-api-key` header
# ANTHROPIC_AUTH_TOKEN= # Alternative to ANTHROPIC_API_KEY; sent as `Authorization: Bearer` header (mutually exclusive)
# ANTHROPIC_BASE_URL=https://your-custom-anthropic/v1
# ANTHROPIC_THINKING_TYPE=enabled # Optional: Anthropic extended thinking (enabled)
# ANTHROPIC_THINKING_BUDGET_TOKENS=12000 # Optional: Budget for extended thinking in tokens
# Google Generative AI Configuration
# GOOGLE_GENERATIVE_AI_API_KEY=...
# GOOGLE_BASE_URL=https://generativelanguage.googleapis.com/v1beta # Optional: Custom endpoint
# GOOGLE_CANDIDATE_COUNT=1 # Optional: Number of candidates to generate
# GOOGLE_TOP_K=40 # Optional: Top K sampling parameter
# GOOGLE_TOP_P=0.95 # Optional: Nucleus sampling parameter
# Note: Gemini 2.5/3 models automatically enable reasoning display (includeThoughts: true)
# GOOGLE_THINKING_BUDGET=8192 # Optional: Gemini 2.5 thinking budget in tokens (for more/less thinking)
# GOOGLE_THINKING_LEVEL=high # Optional: Gemini 3 thinking level (low/high)
# Google Vertex AI Configuration (Enterprise GCP)
# For enterprise users needing data residency, VPC Service Controls, or GCP integration
# GOOGLE_VERTEX_API_KEY= # Required: Express Mode API key
# GOOGLE_VERTEX_BASE_URL=https://... # Optional: Custom endpoint URL
# Note: Gemini 2.5/3 models automatically enable reasoning display (includeThoughts: true)
# GOOGLE_VERTEX_THINKING_BUDGET=8192 # Optional: Gemini 2.5 thinking budget in tokens (1024-100000)
# GOOGLE_VERTEX_THINKING_LEVEL=high # Optional: Gemini 3 thinking level (minimal/low/medium/high)
# Azure OpenAI Configuration
# Configure endpoint using ONE of these methods:
# 1. AZURE_RESOURCE_NAME - SDK constructs: https://{name}.openai.azure.com/openai/v1{path}
# 2. AZURE_BASE_URL - SDK appends /v1{path} to your URL
# If both are set, AZURE_BASE_URL takes precedence.
# AZURE_RESOURCE_NAME=your-resource-name
# AZURE_API_KEY=...
# AZURE_BASE_URL=https://your-resource.openai.azure.com/openai # Alternative: Custom endpoint
# AZURE_REASONING_EFFORT=low # Optional: Azure reasoning effort (low, medium, high)
# AZURE_REASONING_SUMMARY=detailed
# Ollama Configuration (Local or Cloud)
# OLLAMA_BASE_URL=https://ollama.com/api # Optional, defaults to Ollama Cloud
# OLLAMA_API_KEY=your-ollama-cloud-api-key # Optional: For Ollama Cloud or authenticated remote instances
# OLLAMA_ENABLE_THINKING=true # Optional: Enable thinking for models that support it (e.g., qwen3)
# OpenRouter Configuration
# OPENROUTER_API_KEY=sk-or-v1-...
# OPENROUTER_BASE_URL=https://openrouter.ai/api/v1 # Optional: Custom endpoint
# AIHubMix Configuration
# AIHUBMIX_API_KEY=your-aihubmix-api-key
# AIHUBMIX_BASE_URL=https://aihubmix.com/v1 # Optional: Custom endpoint
# DeepSeek Configuration
# DEEPSEEK_API_KEY=sk-...
# DEEPSEEK_BASE_URL=https://api.deepseek.com/v1 # Optional: Custom endpoint
# SiliconFlow Configuration (OpenAI-compatible)
# Base domain can be .com or .cn, defaults to https://api.siliconflow.com/v1
# SILICONFLOW_API_KEY=sk-...
# SILICONFLOW_BASE_URL=https://api.siliconflow.com/v1 # Optional: switch to https://api.siliconflow.cn/v1 if needed
# SGLang Configuration (OpenAI-compatible)
# SGLANG_API_KEY=your-sglang-api-key
# SGLANG_BASE_URL=http://127.0.0.1:8000/v1 # Your SGLang endpoint
# ModelScope Configuration
# MODELSCOPE_API_KEY=ms-...
# MODELSCOPE_BASE_URL=https://api-inference.modelscope.cn/v1 # Optional: Custom endpoint
# ByteDance Doubao Configuration (via Volcengine)
# DOUBAO_API_KEY=your-doubao-api-key
# DOUBAO_BASE_URL=https://ark.cn-beijing.volces.com/api/v3 # ByteDance Volcengine endpoint
# Vercel AI Gateway Configuration
# Get your API key from: https://vercel.com/ai-gateway
# Model format: "provider/model" e.g., "openai/gpt-4o", "anthropic/claude-sonnet-4-5"
# AI_GATEWAY_API_KEY=...
# AI_GATEWAY_BASE_URL=https://your-custom-gateway.com/v1/ai # Optional: Custom Gateway URL (for local dev or self-hosted Gateway)
# # If not set, uses Vercel default: https://ai-gateway.vercel.sh/v1/ai
# Langfuse Observability (Optional)
# Enable LLM tracing and analytics - https://langfuse.com
# LANGFUSE_PUBLIC_KEY=pk-lf-...
# LANGFUSE_SECRET_KEY=sk-lf-...
# LANGFUSE_BASEURL=https://cloud.langfuse.com # EU region, use https://us.cloud.langfuse.com for US
# Optional server-side multi-model configuration
# If set, points to a JSON file with server-provided models (see README for schema).
# Default: ./ai-models.json in project root
# AI_MODELS_CONFIG_PATH=/path/to/ai-models.json
# Temperature (Optional)
# Controls randomness in AI responses. Lower = more deterministic.
# Leave unset for models that don't support temperature (e.g., GPT-5.1 reasoning models)
# TEMPERATURE=0
# Access Control (Optional)
# ACCESS_CODE_LIST=your-secret-code,another-code
# Admin Panel (Optional)
# Set a password to enable the web admin panel at /admin, where most of the
# settings in this file can be edited at runtime (stored in data/settings.json,
# which takes precedence over environment variables).
# Leave unset to disable the admin panel entirely.
# ADMIN_PASSWORD=your-admin-password
# SETTINGS_FILE=./data/settings.json # Optional: custom settings file location
# Draw.io Configuration (Optional)
# NEXT_PUBLIC_DRAWIO_BASE_URL=https://embed.diagrams.net # Default: https://embed.diagrams.net
# Use this to point to a self-hosted draw.io instance
# Subdirectory Deployment (Optional)
# For deploying to a subdirectory (e.g., https://example.com/nextaidrawio)
# Set this to your subdirectory path with leading slash (e.g., /nextaidrawio)
# Leave empty for root deployment (default)
# NEXT_PUBLIC_BASE_PATH=/nextaidrawio
# PDF Input Feature (Optional)
# Enable PDF file upload to extract text and generate diagrams
# Enabled by default. Set to "false" to disable.
# ENABLE_PDF_INPUT=true
# NEXT_PUBLIC_MAX_EXTRACTED_CHARS=150000 # Max characters for PDF/text extraction (default: 150000)
# Security Settings (Optional)
# Allow private/internal URLs for reverse proxy setups (default: true)
# Set to "false" to block private IPs, localhost, and internal hostnames
# ALLOW_PRIVATE_URLS=false
# Self-hosted deployment (Optional)
# Self-hosted users may implement custom quota-management solutions,
# which triggers the client UI to display messages suggesting self-hosting or sponsorship.
# This switch allows self-hosted users to provide custom messages in response to a 429 code,
# in messageTokenSelfHosted, messageApiSelfHosted, and tipSelfHosted translation strings.
# NEXT_PUBLIC_SELFHOSTED=true
# Minimax Configuration (Optional)
# Get your API key from: https://platform.minimaxi.com/docs/guides/models-intro
# MINIMAX_API_KEY=your_minimax_api_key
# MINIMAX_BASE_URL=https://api.minimaxi.com/anthropic # Optional, default (China mainland)
# GLM Configuration (Optional)
# Get your API key from: https://open.bigmodel.cn/dev/api
# GLM_API_KEY=your_glm_api_key
# GLM_BASE_URL=https://open.bigmodel.cn/api/paas/v4 # Optional, default
# Qwen Configuration (Optional)
# Get your API key from: https://www.aliyun.com/product/bailian
# QWEN_API_KEY=your_qwen_api_key
# QWEN_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1 # Optional, default
# Kimi Configuration (Optional)
# Get your API key from: https://platform.moonshot.cn/
# KIMI_API_KEY=your_kimi_api_key
# KIMI_BASE_URL=https://api.moonshot.cn/v1 # Optional, default
# Qiniu Configuration (Optional)
# Get your API key from: https://www.qiniu.com/ai/models
# QINIU_API_KEY=your_qiniu_api_key
# QINIU_BASE_URL=https://api.qnaigc.com/v1 # Optional, default
# Novita AI Configuration (Optional)
# Get your API key from: https://novita.ai/dashboard/key
# NOVITA_API_KEY=your_novita_api_key
# NOVITA_BASE_URL=https://api.novita.ai/openai # Optional, default
# MiMo (Xiaomi) Configuration (Optional)
# Get your API key from: https://platform.xiaomimimo.com/
# MIMO_API_KEY=your_mimo_api_key
# MIMO_BASE_URL=https://api.xiaomimimo.com/v1 # Optional, default. Token Plan users: https://token-plan-cn.xiaomimimo.com/v1
# Atlas Cloud Configuration (Optional)
# Get your API key from: https://www.atlascloud.ai/console/api-keys
# ATLASCLOUD_API_KEY=your_atlascloud_api_key
# ATLASCLOUD_BASE_URL=https://api.atlascloud.ai/v1 # Optional, default. LLM chat endpoint; media generation uses a separate API.
Reference in New Issue View Git Blame Copy Permalink