How long does the integration take?

Change one base URL in an OpenAI-compatible client, or install the Python or TypeScript SDK; the quickstart routes a first request in a few minutes.

Will this lock us into another vendor?

No. The router is Apache-2.0 open source, your provider keys stay yours, and the API is OpenAI-compatible, so leaving is a change of base URL.

What happens when a provider fails?

The router falls back to the next model your policy allows and records the fallback in the trace, so an outage costs a retry, not an incident.

gpuma - Skill - OpenSmartRoute

Imported from niklashoelter/gpuma (AGENTS.md). Install upstream with npx skills add niklashoelter/gpuma. Copyright stays with the author.

AGENTS.md

Guidance for AI coding agents working on the GPUMA repository. (Agents that use GPUMA get the skill in src/gpuma/skills/gpuma/ instead.)

Layout

Path	Content
`src/gpuma/api.py`, `src/gpuma/__init__.py`	Public Python API. Don't change signatures without the maintainer's agreement.
`src/gpuma/cli.py`	argparse CLI; handlers return a `CommandResult`
`src/gpuma/cli_report.py`	Exit codes and the `--json` summary
`src/gpuma/config.py`, `config_schema.py`	Config defaults, validation, JSON Schema
`src/gpuma/optimizer.py`, `autobatching.py`	ASE and torch-sim optimization, batching
`src/gpuma/models/`	Model backends (Fairchem UMA, ORB-v3, SevenNet) and device handling
`src/gpuma/conformer_generation/`	SMILES embedding (nvMolKit on GPU, morfeus on CPU), RMSD pruning
`src/gpuma/utils/reporting.py`	Opt-in diagnostic events used by `--json`
`src/gpuma/doctor.py`, `agent_skills.py`	`gpuma doctor` and `gpuma skills install`
`src/gpuma/skills/gpuma/`	Agent skill shipped to end users
`docs/`	mkdocs site

Environment and tests

uv-managed, Python 3.12 only. Run tests with .venv/bin/python -m pytest. Avoid uv run while someone else's job may hold the project environment.
Tests use real models (ORB by default) and real packages. Only the CLI tests fake the optimization API. GPU tests skip without CUDA; Fairchem tests need HF_TOKEN.
tests/test_skill_content.py fails when the skill names a subcommand, flag or config key that doesn't exist. After renaming any of these, update the skill.
Lint the files you change with ruff check and ruff format (line length 100). Some untouched files have older lint findings; leave those alone.
Docs: mkdocs build --strict must pass.

Conventions

Keep docstrings and comments short; match the surrounding file's idiom.
Fail loudly: an invalid value is an error with the allowed values in the message, never a silent fallback.
The --json summary is a contract for agents. Only add keys; bump SUMMARY_SCHEMA_VERSION in cli_report.py for anything else.
No backward-compatibility shims unless asked for.
Don't add AI co-author trailers or "generated with" footers to commits or PRs.

Install gpuma and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

gpuma

About

AGENTS.md

Layout

Environment and tests

Conventions

Install gpuma and it becomes one more candidate the router can pick - when it fits.

What it costs to adopt - in engineering time.

One base URL

No vendor lock-in

Failover built in

Use it

Manifest

Reviews