botocore

botocore is Amazon Web Services’ low-level Python interface to virtually every AWS service. Rather than hand-writing a client for each service, botocore generates clients from JSON service models, giving it broad and immediately up-to-date coverage as AWS ships new APIs and services.

It is not typically used directly by application developers; instead it is the shared foundation that boto3 (the higher-level, resource-oriented SDK) and the official AWS CLI are both built on. Anyone building session handling, credential resolution, request signing, or a custom AWS tool at a lower level than boto3 provides depends on botocore directly.

What You Get

• A botocore.session.get_session() entry point that creates low-level clients for any AWS service (session.create_client('ec2'), etc.)
• Automatic credential resolution across environment variables, shared config/credentials files, IAM roles, and SSO
• AWS Signature Version 4 request signing and endpoint resolution handled transparently per service/region
• Service clients generated from bundled JSON API models, so new AWS services and operations become available as botocore is updated rather than requiring hand-written bindings
• Optional awscrt-backed extras for CRT-accelerated signing/checksums on performance-sensitive workloads

Common Use Cases

• Serving as the dependency that boto3 and the AWS CLI use internally for every AWS API call
• Building a custom, lightweight AWS tool that needs direct client/session control without boto3’s resource abstractions
• Implementing custom credential providers or request signing/middleware hooks for AWS API calls via botocore’s event system
• Powering third-party libraries (e.g. cloud storage adapters, infrastructure tools) that need AWS API access without the full boto3 surface

Under The Hood

Architecture: The core loop runs through session.py (Session/credential/config orchestration), client.py (dynamic per-service client class generation from model data), model.py (parses the JSON service models under data/ into a queryable ServiceModel/OperationModel structure), endpoint.py/endpoint_provider.py (region/partition endpoint resolution), auth.py (SigV4/SigV4A request signing), and hooks.py (the pluggable event system other AWS Python tools, including boto3, hook into to customize behavior). Tech Stack: Pure Python with a small dependency set (jmespath for response filtering, python-dateutil, urllib3), plus an optional crt extra pulling in awscrt for native-accelerated signing/checksums; the data/ directory bundles versioned JSON service models for every supported AWS service. Code Quality: Ruff-linted with a strict 79-char line length (Black-compatible), a pytest suite with dedicated slow and validates_models markers (the latter validating the bundled service models themselves), and a well-documented CONTRIBUTING process reflecting Amazon’s ownership and formal SDK maintenance policy. API Design: The session.create_client('service-name') pattern is uniform across all ~300+ AWS services since clients are generated from the same model schema, but the API is intentionally low-level (raw dict request/response shapes, manual pagination in some cases) — developers wanting resource objects, higher-level abstractions, or waiters typically reach for boto3 instead, which layers those on top of botocore’s client objects.