benchflow/.env.sample at main · benchflow-ai/benchflow · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
# BenchFlow environment variables
# Copy to .env and fill in your keys: cp .env.sample .env

# ── Sandbox ──────────────────────────────────────────────────
# Docker works out of the box. For cloud sandboxes, set Daytona keys:
# DAYTONA_API_KEY=
# DAYTONA_API_URL=https://app.daytona.io/api

# ── API keys (set the ones you need) ─────────────────────────
# If you've logged in with the agent CLI (claude login, gemini auth,
# codex --login), BenchFlow auto-detects the host credentials and
# API keys below are optional. Explicit keys always take precedence.
#
# Auth precedence:
#   1. Vertex ADC (gcloud auth application-default login)
#   2. Explicit API key below (or via --ae) — includes provider keys like ZAI_API_KEY
#   3. Host CLI subscription auth (~/.claude/.credentials.json, etc.)
#
# ANTHROPIC_API_KEY=          # claude-agent-acp, pi-acp
# OPENAI_API_KEY=             # codex-acp
# GEMINI_API_KEY=             # gemini
# ZAI_API_KEY=                # zai/ provider (openclaw, claude-agent-acp)

# ── Provider options ────────────────────────────────────────
# Optional OpenAI prompt cache retention policy. Allowed: in_memory, 24h.
# BENCHFLOW_PROVIDER_PROMPT_CACHE_RETENTION=24h

# ── AWS Bedrock (API key auth, optional) ────────────────────
# Minimum required for Bedrock proxy backend:
# AWS_BEARER_TOKEN_BEDROCK=
# AWS_REGION=us-east-1
# CLAUDE_CODE_USE_BEDROCK=1
# ANTHROPIC_DEFAULT_SONNET_MODEL='us.anthropic.claude-sonnet-4-5-20250929-v1:0'


# ── Google Cloud (for Vertex AI providers) ───────────────────
# Requires: gcloud auth application-default login
# GOOGLE_CLOUD_PROJECT=
# GOOGLE_CLOUD_LOCATION=global