distillr/.env.example at main · blisspixel/distillr · 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
# Distill API Keys
# Get XAI key at: https://console.x.ai/
XAI_API_KEY=

# Gemini analysis route and Deep Research reports: https://aistudio.google.com/apikey
# After init (xAI default), switch analysis with:
#   distill provider set gemini gemini-3.6-flash
GEMINI_API_KEY=

# Optional: OpenAI key (not used in main flow)
OPENAI_API_KEY=

# Optional: Path to scribe for audio-only transcription fallback
# SCRIBE_PATH=

# Optional: Override defaults
# DISTILL_OUTPUT_DIR=./library
# DISTILL_DEFAULT_MONTHS=1
# Cost policy: auto | no-metered | paid-ok. no-metered allows local routes
# and blocks API-billed or ambiguous provider routes until they have adapter
# proof.
# DISTILL_COST_MODE=auto
# Optional direct CLI workflow spend caps in USD. Estimate-bearing commands
# refuse before known model work; other tracked commands stop after the crossing
# call is recorded.
# DISTILL_COST_WORKFLOW_BUDGETS=ask=0.25,report=5,discover=2,eval=1,ingest=1,paper=1,papers=2,video=1,channel=2,catch-up=2,reanalyze=2,resynthesize=1,site=3,site-batch=3,corpus=1,topic-brief=1,synthesize=1,synthesis=1

# ─── Provider Selection ───────────────────────────────────────────────
# Default: xai (cloud). Prefer `distill provider set <provider> [model]`.
# DISTILL_PROVIDER=xai
# DISTILL_PROVIDER=gemini
# DISTILL_PROVIDER=ollama
# DISTILL_PROVIDER=lmstudio

# Local provider URLs (defaults shown)
# OLLAMA_BASE_URL=http://localhost:11434
# LMSTUDIO_BASE_URL=http://localhost:1234/v1
# Maximum seconds without a streamed local token, and the total bound for waiting
# on a different resident Ollama model. Progressing inference can run longer.
# A contention timeout returns retry guidance and never substitutes models.
# DISTILL_LOCAL_TIMEOUT=1800

# Per-workload provider overrides (e.g., use cloud for synthesis, local for analysis)
# DISTILL_ANALYSIS_PROVIDER=ollama
# DISTILL_SYNTHESIS_PROVIDER=xai
# DISTILL_REPORT_PROVIDER=xai

# ─── Model Selection ─────────────────────────────────────────────────
# Cloud tier defaults are grok-4.3. Local routes require an exact model id.
# DISTILL_MODEL=qwen3.5:27b
# DISTILL_MODEL=gemini-3.6-flash
# XAI_FAST_MODEL=grok-4.3
# XAI_PREMIUM_MODEL=grok-4.3
# XAI_SITE_MODEL=grok-4.3
# ACCORDION_SECTION_MODEL=grok-4.3

# Per-workload model overrides (works with any provider)
# DISTILL_ANALYSIS_MODEL=qwen3.5:27b
# DISTILL_RERANK_MODEL=qwen3.5:27b
# DISTILL_SYNTHESIS_MODEL=qwen3.5:27b

# ─── Reasoning Effort ────────────────────────────────────────────────
# Grok 4.3 valid: low, medium, high. Premium workloads default to "high",
# fast workloads default to "medium".
# Claude Sonnet 5 valid: low, medium, high, xhigh, max. Distill forwards it
# only when explicitly set because Sonnet 5 adaptive thinking is on by default.
# DISTILL_ANALYSIS_REASONING_EFFORT=medium
# DISTILL_SITE_REASONING_EFFORT=high
# DISTILL_REPORT_REASONING_EFFORT=high

# ─── MCP server posture (agent-facing deployments) ───────────────────
# Read-only: agents keep the full read surface; every spend/ingest/mutation
# tool refuses. The recommended posture when agents you don't control connect.
# DISTILL_MCP_READ_ONLY=1
# If write tools ARE exposed: cap each tool call's recorded spend (dollars),
# and confine URL ingest to approved hosts (+ subdomains).
# DISTILL_MCP_MAX_SPEND_PER_CALL=0.50
# DISTILL_MCP_INGEST_ALLOWLIST=youtube.com,learn.microsoft.com