generic-ml-cache-core 0.8.0

Hexagonal core library for generic-ml-cache: domain, use cases, ports, and the default outbound adapters (SQLite repo, blob store, local clients, API). Stateless; inject the data source.

generic-ml-cache-core

The hexagonal engine behind gmlcache — embeddable and stateless

The reusable engine behind gmlcache: record a real ML client (or API) call once, replay it by its content key. It contains the domain model, the use cases, the port contracts, and the default outbound adapters (SQLite execution repository, filesystem blob store, the claude/codex/cursor client runner, the API client, metrics, clock, fingerprinting) — plus the build_use_cases composition factory.

Pure Python and stateless: it bakes in structure (table names, blob naming, schema) but no location — you inject the data source.

Part of a single-user, local tool — not a gateway. gmlcache records and replays across the subscriptions and APIs you already hold; it is not a multi-user router. See Positioning.

Install

pip install generic-ml-cache-core

Embed it

Hand the library a data source and it wires the engine for you:

from generic_ml_cache_core import build_use_cases
from generic_ml_cache_core.application.port.inbound.run_managed_local_execution_command import (
    RunManagedLocalExecutionCommand,
)

wired = build_use_cases(store_root="/path/you/choose")   # you provide the data source
command = RunManagedLocalExecutionCommand(
    client="claude", model="sonnet", effort="", context="", prompt="…",
)
execution = wired.run_managed.execute(command)           # records on a miss, replays on a hit

You reuse the shipped adapters by injecting a data source — you never reimplement them (the Spring Batch model: the framework ships the writers, you provide the connection). Need a different store? Construct the use cases yourself against the ports and pass your own adapter.

What's inside

• Domain model — executions, polymorphic call identities, artifacts, usage.
• Use cases — managed-local / passthrough / API runs, and probe (check).
• Ports (application/port/...) — client runner, blob store, execution repository, metrics, clock, fingerprint, API client.
• Default adapters (adapter/out/...) + the build_use_cases composition factory.
• generic_ml_cache_core.testing.InMemoryExecutionRepository — an in-memory reference adapter to test your code against the ports.

Inbound drivers — gmlcache today, a daemon later — map their surface (a terminal, a REST API) onto these public APIs; the core itself has no UI and reads no config file.

Links

• Repository & docs: https://github.com/danielslobozian/generic-ml-cache
• Changelog (both packages, versioned in lockstep): CHANGELOG.md
• Security policy: SECURITY.md

License

Apache-2.0 — see LICENSE and NOTICE.