Configuration

Configuration file

Create baseline.config.ts in your project root:

import type { BaselineConfig } from 'baselineos';

export default {
  projectRoot: '.',
  name: 'my-project',
  knowledge: ['./docs', './src'],
  llm: {
    provider: 'anthropic',
    model: 'claude-sonnet-4-6',
  },
  orchestrator: {
    defaultTrustScore: 70,
    selfVerificationRequired: true,
    qualityReviewSampleRate: 0.1,
  },
  governance: {
    auditLevel: 'full',
    requireApproval: ['critical'],
    restricted: [],
  },
} satisfies Partial<BaselineConfig>;

Key options

knowledge

An array of filesystem paths to index. Baseline scans these directories for docs, code, configs, and specs to build its knowledge graph.

llm

Configure the AI provider and model. Baseline supports multiple providers — see the multi-LLM neutrality design. The ANTHROPIC_API_KEY environment variable is required when using Anthropic models.

orchestrator

• defaultTrustScore — Initial trust score for new agents (0-100)
• selfVerificationRequired — Whether agents must self-verify before review
• qualityReviewSampleRate — Fraction of tasks sent to quality review (0.0-1.0)

governance

• auditLevel — full records every decision; summary records outcomes only
• requireApproval — Priority levels that require human approval before execution
• restricted — Operations that are never allowed autonomously

Environment variables

Variable	Purpose
ANTHROPIC_API_KEY	Required for LLM-powered execution
BASELINE_OTEL_DISABLED=1	Suppress OpenTelemetry noise in local dev
BASELINE_APPROVAL_SECRET	Optional cryptographic approval token for task approval

JSON Schemas

For non-TypeScript implementations, JSON Schemas are available:

• specs/schemas/protocol-core.schema.json — Protocol types
• specs/schemas/govern.schema.json — Governance types
• specs/schemas/evidence-bundle.schema.json — Evidence bundle format
• specs/schemas/orchestrator.schema.json — Task and agent types