Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/metareflection/claimcheck/llms.txt

Use this file to discover all available pages before exploring further.

Claimcheck delegates all LLM work to Claude models. By default it calls the Anthropic API directly, but you can route requests through Vertex AI, AWS Bedrock, or the Claude Code CLI instead. This page covers how to configure each backend and how to override the default models used for informalization and comparison.

​
Default backend: Anthropic API

When no backend flag is provided, Claimcheck uses the Anthropic API with your ANTHROPIC_API_KEY environment variable. 
export ANTHROPIC_API_KEY=sk-ant-...
claimcheck -m mappings.json --dfy claims.dfy -d counter
ANTHROPIC_API_KEY must be set in your environment. Claimcheck does not read .env files.

​
Vertex AI (--vertex)

Pass --vertex to route requests through the Anthropic Vertex AI SDK. You must supply a Google Cloud project ID and optionally a region. 
claimcheck \
  -m mappings.json --dfy claims.dfy -d counter \
  --vertex --vertex-project my-gcp-project
​
--vertex-project
string
Google Cloud project ID. Can also be set via the GOOGLE_CLOUD_PROJECT_ID environment variable.
​
--vertex-region
string
default:"us-east5"
Google Cloud region. Defaults to us-east5 when not provided. Can also be set via GOOGLE_CLOUD_REGION.
Vertex AI uses shorter model names (no date suffix). See the model defaults table below.

​
AWS Bedrock (--bedrock)

Pass --bedrock to route requests through the Anthropic Bedrock SDK. Claimcheck uses the standard AWS credential chain — environment variables, ~/.aws/credentials, SSO, or IMDS — so no extra configuration is required beyond ensuring your credentials are in place. 
claimcheck \
  -m mappings.json --dfy claims.dfy -d counter \
  --bedrock
​
--bedrock-region
string
default:"us-east-1"
AWS region for Bedrock. Defaults to us-east-1. Can also be set via the AWS_REGION environment variable.
Default Bedrock model IDs use the us. cross-region inference profile prefix:
  • Informalize step: us.anthropic.claude-haiku-4-5-20251001-v1:0
  • Compare step: us.anthropic.claude-sonnet-4-6
Bedrock credentials are resolved in the standard AWS order: environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY), shared credentials file (~/.aws/credentials), IAM role via SSO, and instance metadata (IMDS). No Claimcheck-specific credential configuration is needed.

​
Claude Code CLI (--claude-code)

Pass --claude-code to route requests through the claude -p CLI instead of any direct API. This uses short model aliases (haiku, sonnet) and does not require an API key to be configured in your shell — the Claude Code CLI manages authentication. 
claimcheck \
  -m mappings.json --dfy claims.dfy -d counter \
  --claude-code
If you are running Claimcheck from inside a Claude Code session, the claude -p subprocess inherits the same terminal. Redirect output to files to avoid interleaving:
claimcheck -m mappings.json --dfy claims.dfy -d counter \
  --claude-code --json > results.json 2> claimcheck.log

​
Model defaults

Each backend has its own set of default model IDs. The informalization step (two-pass mode) uses the haiku model; the comparison step and single-prompt mode use the sonnet model.
BackendStepDefault model ID
api (Anthropic)haiku (informalize)claude-haiku-4-5-20251001
api (Anthropic)sonnet (compare)claude-sonnet-4-5-20250929
vertexhaiku (informalize)claude-haiku-4-5
vertexsonnet (compare)claude-sonnet-4-6
bedrockhaiku (informalize)us.anthropic.claude-haiku-4-5-20251001-v1:0
bedrocksonnet (compare)us.anthropic.claude-sonnet-4-6
cc (Claude Code)haiku (informalize)haiku
cc (Claude Code)sonnet (compare)sonnet

​
Overriding models

Use the following flags to override any default model ID for any backend.
​
--model
string
Override the model used for the comparison step (two-pass mode) or the single combined step (single-prompt mode).
​
--informalize-model
string
Override the model used for the informalization (back-translation) step in two-pass mode. Has no effect in single-prompt mode.
​
--compare-model
string
Override the model used for the comparison step in two-pass mode. Has no effect in single-prompt mode.

​
Examples

claimcheck -m mappings.json --dfy claims.dfy -d counter \
  --compare-model claude-opus-4-5-20251001
When using the library API, pass the same overrides via the options object: 
import { claimcheck } from 'claimcheck';

const { results } = await claimcheck({
  claims,
  domain: 'counter',
  options: {
    vertex: true,
    informalizeModel: 'claude-haiku-4-5',
    compareModel: 'claude-sonnet-4-6',
  },
});

Build docs developers (and LLMs) love

Get started for freeTalk to us