AWS Bedrock adapter

Route ABS llm_judge and custom evaluations through an Amazon Bedrock model.

Routes ABS llm_judge and custom evaluations through an Amazon Bedrock model (Converse API). No spans, no AgentCore, no agent instrumentation.

Availability. Ships with the Python CLI and the npm CLI. On npm, install @aws-sdk/client-bedrock-runtime; credentials use the standard AWS chain.

Install

pip install "abslang[aws]"

Configure

export AWS_REGION="us-east-1"
export AWS_PROFILE="my-profile"   # optional
export BEDROCK_EVALUATOR_MODEL_ID="anthropic.claude-3-5-haiku-20241022-v1:0"

Run

abslang run session.abs.yaml --agent $AGENT_URL --adapter aws

Supported evaluators

ABS	What it does
`llm_judge`	Bedrock model scores the response 0–1 against `criteria`
`custom`	User-defined metric via `prompt` + `rating_scale`

custom supports a rating scale (1-5, 0-1) and placeholders {{response}}, {{query}}, {{context}}, {{criteria}}, {{trace}}. The adapter normalizes the returned score to 0–1 before threshold is applied.