About

Arc is a vendor-neutral technology radar for applied AI: a catalog of models, tools, techniques, and infrastructure for teams building production systems on LLMs and agents. We aren't tied to any single vendor — proprietary and open solutions sit side by side, managed services next to self-hosted, mature defaults next to promising newcomers. Every entry is placed in one of four quadrants (Models, Tools, Techniques, Infrastructure) and one of four maturity rings. The goal is to give an engineer a fast, honest bearing: what you can already build on, what's worth piloting, what to keep watching, and what to steer clear of.

Methodology

A ring reflects not "quality" in the abstract but how ready a technology is for responsible adoption: how well proven it is under real load, how predictable it is to operate, how stable its API is, and how well it fits the standards. We weigh maturity and support, availability (open-weight/open-source vs managed), results on reproducible benchmarks, operational cost, and the presence of a living ecosystem. The radar is deliberately vendor-neutral: neighboring entries in the same ring are alternatives to compare, not a better/worse ranking. Placement reflects the state of the industry on the revision date and is revisited as releases ship, statuses change (GA, maintenance, archival), and production experience accumulates; stale or wound-down projects move to Hold, and matured ones are promoted up the rings.

Maturity rings

Adopt. A mature default. The technology is proven under real load, predictable to operate, and has a settled API — safe to build production and new projects on with confidence.

Trial. Ready for pilots. Production-viable and delivers tangible value, but takes real adoption effort, tuning to your profile, or is still maturing — trial it on bounded tasks with good evals before making it a default.

Assess. Worth watching. A promising direction with a fast-maturing ecosystem or a fresh release; governance, security, and operational patterns are still forming — explore it in isolated scenarios and behind opt-in, not trusting it with critical production.

Hold. Hold off. A stalled, wound-down, or successor-superseded option. Not a sensible pick for new systems; maintain existing deployments and migrate off in an orderly way.

Updates

The radar is revisited as releases ship and production experience accumulates. A machine-readable version lives at /radar.json.