DrugDiscoveryBench/.env.example at main · scaleapi/DrugDiscoveryBench · 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
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
# Biomni environment configuration.
# Copy to `.env` and fill in the values you need: `cp .env.example .env`
# Every secret is read from the environment. Only set the keys you actually use.

# =============================================================================
# LLM provider credentials — set the one(s) matching BIOMNI_SOURCE / BIOMNI_LLM
# =============================================================================
# Anthropic (Claude)
ANTHROPIC_API_KEY=

# OpenAI (and OpenAI-compatible endpoints)
OPENAI_API_KEY=

# Google Gemini
GEMINI_API_KEY=

# Groq
GROQ_API_KEY=

# Custom / self-hosted / LiteLLM-compatible proxy (used when BIOMNI_SOURCE=Custom)
# BIOMNI_CUSTOM_BASE_URL=https://your-proxy.example.com/v1
# BIOMNI_CUSTOM_API_KEY=

# LLM selection (non-secret)
# BIOMNI_SOURCE=Anthropic
# BIOMNI_LLM=claude-sonnet-4-6

# =============================================================================
# Benchmark eval credentials (scripts/run_eval.sh) — see docs/model-proxy.md
# =============================================================================
# Judge (verifier) — any OpenAI-API-compatible endpoint. task.toml reads these;
# JUDGE_MODEL defaults to gpt-4o if unset. The model can be from any provider as
# long as JUDGE_BASE_URL speaks the OpenAI format.
# JUDGE_BASE_URL=https://api.openai.com/v1
# JUDGE_API_KEY=
# JUDGE_MODEL=gpt-4o

# Single-proxy shortcut: set these and run_eval.sh derives the per-role agent +
# judge keys/URLs (any per-role var set above/explicitly still wins).
# LLM_API_KEY=
# LLM_BASE_URL=https://your-proxy.example.com

# =============================================================================
# Server authentication (optional — recommended for any non-local deployment)
# =============================================================================
# Bearer token required on protected endpoints. Unset = auth disabled (local dev only).
# BIOMNI_SCALE_API_KEY=
# Bearer token for the /prompt eval endpoint. Unset = unauthenticated.
# BIOMNI_EVAL_API_KEY=

# =============================================================================
# Tool API keys / identifiers (optional — enable richer results for these tools)
# =============================================================================
# Web search (Brave)
# BRAVE_SEARCH_API_KEY=
# Literature (Semantic Scholar)
# SEMANTIC_SCHOLAR_API_KEY=
# Synapse — set to access private studies; public catalog works without it
# SYNAPSE_AUTH_TOKEN=
# protocols.io access token (either name works)
# PROTOCOLS_IO_ACCESS_TOKEN=
# BIOMNI_PROTOCOLS_IO_ACCESS_TOKEN=
# NCBI Entrez identification email (not secret)
# NCBI_EMAIL=you@example.com

# =============================================================================
# Observability (optional)
# =============================================================================
# Datadog metrics — no-op unless set
# DD_API_KEY=

# =============================================================================
# Tool-output capture cache (optional)
# NOTE: in this codebase capture turns on whenever BIOMNI_CAPTURE_TOOL_DATA is
# set, and BIOMNI_CAPTURE_BUCKET currently DEFAULTS to an internal bucket. To
# avoid writing to that default, leave capture off, or set your own bucket.
# =============================================================================
# BIOMNI_CAPTURE_TOOL_DATA=1
# BIOMNI_CAPTURE_BUCKET=your-s3-bucket
# BIOMNI_CAPTURE_PREFIX=tool-captures
# BIOMNI_CAPTURE_MAX_BYTES=256000000

# Artifact upload (plots/CSVs/etc.) — "none" (default) keeps artifacts inline/local
# BIOMNI_ARTIFACT_STORAGE=none
# BIOMNI_ARTIFACT_S3_BUCKET=

# Deployed BiOMNI endpoint for remote eval runs (eval_pass_at_k.py --remote)
BIOMNI_REMOTE_URL=https://your-biomni-endpoint.example.com