vectorless-engine/config.example.yaml at main · hallelx2/vectorless-engine · 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
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
# vectorless-engine example configuration.
#
# Three ways to configure the engine, in increasing priority:
#   1. YAML file   - copy this to config.yaml and edit.
#   2. Environment - VLE_* vars, dot-path flattened with underscores
#                    (e.g. server.tls.cert_file -> VLE_SERVER_TLS_CERT_FILE).
#   3. CLI flags   - mirror the YAML tree, dot-separated
#                    (e.g. --server.addr=:8080, --log.level=debug).
#
# Later layers override earlier ones on a per-key basis (not per-subtree),
# so you can ship a YAML file in the image, set secrets via env, and tweak
# individual knobs per run with flags.
#
# `vectorless-engine config print` prints the effective config with secrets
# redacted; `vectorless-engine config check` validates it and exits 0/1.

server:
  addr: ":8080"
  read_timeout: 30s
  write_timeout: 120s

  # TLS is OPT-IN. Leave both files empty to serve plaintext HTTP behind a
  # reverse proxy (Caddy, nginx, an ALB, ingress) — the recommended
  # production setup because cert rotation stays in the proxy. Set both
  # cert_file and key_file to have the engine terminate TLS directly
  # (useful for single-node / homelab / MCP-over-public-internet).
  tls:
    cert_file: ""            # path to PEM cert chain
    key_file: ""             # path to PEM private key
    min_version: "1.2"       # "1.2" | "1.3"

database:
  # Postgres connection used for documents, sections (the tree), and (if
  # queue.driver=river) the job queue.
  url: "postgres://vectorless:vectorless@localhost:5432/vectorless?sslmode=disable"
  max_conns: 10

storage:
  # Where document bytes are stored.
  # driver: local | s3
  driver: "local"

  local:
    root: "./data/documents"

  s3:
    # Works for AWS S3, Cloudflare R2, MinIO, Backblaze B2, DigitalOcean Spaces,
    # and any other S3-compatible provider — just point endpoint at their URL.
    endpoint: "http://localhost:9000"
    region: "us-east-1"
    bucket: "vectorless-docs"
    access_key: "minio"
    secret_key: "miniominio"
    use_path_style: true     # true for MinIO / R2 / most non-AWS providers

queue:
  # Where background jobs (ingest, tree-build, summarize) are scheduled.
  # driver: qstash | river | asynq
  driver: "river"

  qstash:
    # Upstash QStash — ideal for serverless hosts (Vercel, Cloudflare Workers).
    token: ""
    webhook_base_url: "https://your-engine.example.com"

  river:
    # Postgres-backed, uses database.url above. No extra infra needed.
    num_workers: 10

  asynq:
    # Redis-backed. Higher throughput when Redis is already available.
    addr: "localhost:6379"
    password: ""
    db: 0
    concurrency: 20

llm:
  # Provider used for tree construction and retrieval reasoning.
  # driver: anthropic | openai | gemini
  driver: "anthropic"

  anthropic:
    api_key: ""
    model: "claude-sonnet-4-5"
    reasoning_model: "claude-opus-4-5"   # optional override for deep-reason strategy

  openai:
    api_key: ""
    model: "gpt-4o-mini"
    reasoning_model: "gpt-4o"

  gemini:
    api_key: ""
    model: "gemini-2.0-flash"
    reasoning_model: "gemini-2.5-pro"

retrieval:
  # strategy: single-pass | chunked-tree
  strategy: "chunked-tree"

  chunked_tree:
    # Max tokens of tree view to feed a single LLM call (per subtree slice).
    max_tokens_per_call: 60000
    # Max parallel LLM calls when the tree must be split.
    max_parallel_calls: 8
    # If true, include summary-only breadcrumbs of sibling subtrees each call
    # doesn't own, so the model knows what else exists in the document.
    include_sibling_breadcrumbs: true

log:
  level: "info"            # debug | info | warn | error
  format: "json"           # json | console