Skip to content

configuration.md

Latest commit

 

History

History
219 lines (177 loc) · 6.04 KB

configuration.md

File metadata and controls

219 lines (177 loc) · 6.04 KB

Configuration Reference

Config File Locations

Quorum searches for configuration in this order:

  1. Project-local: ./quorum.yaml (or legacy ./counsel.yaml)
  2. Global: ~/.quorum/config.yaml

Top-Level Config

# ~/.quorum/config.yaml
providers:
  - name: claude
    provider: anthropic
    model: claude-sonnet-4-20250514
    auth:
      method: env
      envVar: ANTHROPIC_API_KEY
    timeout: 120  # seconds (optional)

  - name: ollama
    provider: ollama
    model: llama3
    auth:
      method: none
    baseUrl: http://localhost:11434  # optional

defaultProfile: default
profiles: {}

Provider Config Fields

Field Type Required Description
name string ✅ Unique identifier for this provider
provider string ✅ Provider type: openai, anthropic, codex, ollama, google, gemini-cli, mistral, deepseek, kimi, custom
model string ✅ Model identifier
auth object — Authentication config (see below)
baseUrl string — Custom API base URL
timeout number — Per-provider timeout in seconds (default: 120)

Auth Methods

# Environment variable (recommended)
auth:
  method: env
  envVar: OPENAI_API_KEY

# Direct API key (not recommended — use env instead)
auth:
  method: api_key
  apiKey: sk-...

# OAuth (Claude Code, etc.)
auth:
  method: oauth
  profileName: claude-code

# macOS Keychain
auth:
  method: oauth_keychain
  service: com.anthropic.claude-code

# No auth (local models)
auth:
  method: none

Agent Profiles

Profiles live in ~/.quorum/agents/<name>.yaml or inline in config.yaml.

name: my-profile
rounds: 1
focus: [security, performance]
challengeStyle: adversarial  # adversarial | collaborative | socratic
isolation: true
blindReview: false

# Phase customization
phases: [gather, debate, adjust, vote, synthesize]  # custom pipeline

# Provider roles
roles:
  claude: "Security expert focused on OWASP Top 10"
  openai: "Performance engineer"

# Voting
votingMethod: borda  # borda | ranked-choice | approval | condorcet
weights:
  claude: 1.5
  openai: 1.0

# Evidence
evidence: strict  # off | advisory | strict

# Adaptive debate
adaptive: balanced  # fast | balanced | critical | off

# Topology
topology: mesh  # mesh | star | tournament | map_reduce | adversarial_tree | pipeline | panel
topologyConfig:
  hub: claude          # for star topology
  moderator: openai    # for panel topology

# Tools
tools: true
allowShellTool: false

# Red team
redTeam: true
attackPacks: [security, code]

# Human-in-the-loop
hitl: true
hitlPhases: [debate, vote]

# Hooks
hooks:
  pre-gather: "./scripts/fetch-context.sh"
  post-synthesis: "./scripts/notify-slack.sh"

# Scoring weights
scoringWeights:
  accuracy: 0.3
  reasoning: 0.25
  completeness: 0.2
  novelty: 0.15
  consensus: 0.1

# Custom phase prompts
prompts:
  gather:
    system: "You are an expert analyst..."
  debate:
    system: "Challenge every assumption..."

All Profile Fields

Field Type Default Description
name string — Profile name
rounds number 1 Number of deliberation rounds
focus string[] [] Topics to focus on
challengeStyle string adversarial How providers critique: adversarial, collaborative, socratic
isolation boolean true Whether providers see each other in gather phase
blindReview boolean false Hide provider names during review
phases string[] all 7 Custom phase pipeline
roles object — Per-provider role descriptions
votingMethod string borda Voting algorithm
weights object — Per-provider vote weights
evidence string off Evidence mode: off, advisory, strict
adaptive string off Adaptive preset: fast, balanced, critical, off
topology string mesh Debate topology
tools boolean false Enable tool use in gather
allowShellTool boolean false Enable shell tool
redTeam boolean false Enable red team analysis
attackPacks string[] — Red team attack pack names
devilsAdvocate boolean false Force one provider contrarian
decisionMatrix boolean false Structured decision scoring
hitl boolean false Enable HITL checkpoints
hooks object — Shell commands per phase
convergenceThreshold number — Score threshold for early consensus
excludeFromDeliberation string[] — Provider names to skip
prompts object — Custom system prompts per phase
scoringWeights object — Scoring dimension weights

Built-in Profiles

Profile Description
default Balanced deliberation
brainstorm Creative, collaborative style
code-review Focused on code quality
research Deep analysis with evidence
decision Structured decision matrix
panel Panel-style with moderator
quick Fast, fewer phases
thorough Maximum depth
evidence Evidence-strict mode
research-tools Research with web search enabled

Policy Files

Policies live in ~/.quorum/policies/ or .quorum/policies/.

name: production
rules:
  - condition: confidence < 0.7
    action: block
    message: "Confidence too low for production decisions"
  - condition: evidence_grade < C
    action: warn
    message: "Evidence quality below threshold"
  - condition: providers_count < 3
    action: pause
    message: "Consider adding more providers"

Actions: block, warn, log, pause. Rules evaluate pre- and post-deliberation.

Project-Local Config

Create .quorumrc in your project root. Quorum walks from cwd to homedir looking for it.

Environment Variables

Variable Description
QUORUM_CONFIG Override config file path
QUORUM_HOME Override config directory (default: ~/.quorum)
Provider API keys See providers.md