CLI and Config
Initialize A Starter
Use groundguard-init to create a minimal config and sample files:
groundguard-init --template github-action
Available templates:
groundguard-init --template github-action
groundguard-init --template openai
groundguard-init --template promptfoo
groundguard-init --template langgraph
groundguard-init --template pydanticai
groundguard-init --template crewai
groundguard-init --template autogen
groundguard-init --template fastapi
The command writes groundguard.yml, groundguard-ledger.jsonl, answer.txt,
and a small template-specific starter file. Existing files are not overwritten
unless you pass --force.
Generate a Report
groundguard-report \
--ledger-jsonl facts.jsonl \
--answer-file answer.txt \
--required-fact revenue_2025 \
--schema assertion \
--fail-on-policy
Use groundguard.yml
Start from the repository sample:
cp groundguard.example.yml groundguard.yml
required_facts:
- revenue_2025
policy:
allow_candidate_matches: false
on_contradicted: block
on_unverified: flag
units:
tolerance: 0.005
report:
schema: assertion
format: json
fail_on_policy: true
groundguard-report --config groundguard.yml --ledger-jsonl facts.jsonl --answer-file answer.txt
CLI flags extend or override the config where appropriate. For example,
additional --required-fact values are appended to required_facts.
Render Human Reports
groundguard-report \
--config groundguard.yml \
--ledger-jsonl facts.jsonl \
--answer-file answer.txt \
--format markdown \
--output groundguard-report.md
Supported formats are json, markdown, html, and github. The github
format is a compact Markdown body designed for PR comments.
Reports include matched_fact_key, ledger_value, answer_value, and
start:end spans for claim highlighting and review UIs.
Scoped Extractor Packs
extractors:
packs:
- finance
- saas
- ops
Extractor packs are resolved per report run. They do not mutate the global process registry, so they are safe to use in multi-tenant services.
Built-in packs cover common labels for finance (revenue, gross_margin,
eps), SaaS (arr, mrr, churn, nrr, cac, ltv), ecommerce (gmv,
orders, conversion_rate, aov), and ops (p95_latency, error_rate,
uptime, throughput, storage).