Sober, evidence-tied read on Nyx's market position, where it differentiates, where it loses, and what to charge. Written by a Nyx research worker (opus 4.8). Not a pitch.
Nyx is a genuinely differentiated piece of software with a thin but real moat: a hardcoded, test-proven safety floor plus signal-driven self-iteration that no shipping competitor matches at this level of safety wiring. The lead pitch for every enterprise conversation is the three-tier Push Gate whose Tier-1 floor cannot be overridden by any operator config (proven by floor-cannot-be-lifted.test.ts) -- this is the answer to "what stops it from going rogue." BYO-Claude-Code-subscription is a first-class feature: teams bring their own Claude seat, Nyx orchestrates it, no token markup, no vendor lock-in. Price it as a per-seat team tool (not a metered cloud agent), because Nyx sells software and safety rather than tokens. Recommended structure: Community free for adoption, Team at $39/user/month for engineering teams, Enterprise at custom for self-hosted deployments with audit log exports and SLA. The system now stands at roughly 31k lines (19k source + 12k tests) across 264 commits, with a token-free status pulse (GET /status/pulse), a self-improvement reflector (GET /reflection/digest), a Verifier runtime gate (in queue), and 5 worker-deadlock root-cause fixes all shipped as of 2026-05-30.
Pro tier monthly prices as of 2026-05-28. Devin shows its base subscription; the cloud autonomy model is fundamentally different. Nyx is the only entry that includes a non-overridable safety floor at the floor price.
Six features that materially separate Nyx from the closest competitors. Check = full support, dot = partial, dash = absent. Most differentiation comes from safety + self-iteration + local-first; the rest is convergent.
Nyx is local-first and BYO-Claude-Code-subscription -- the operator does not resell inference. That means near-zero cost of goods and no lock-in for buyers. Price like a per-seat team license (Cursor seat, JetBrains plan), not a metered agent. Community drives adoption; Team and Enterprise are where the revenue lives.
Why BYO-subscription is a feature, not a constraint: Teams using Claude Max or Pro already pay Anthropic directly. Nyx orchestrates that subscription without touching tokens -- no per-ACU markup, no vendor dependency. This is the strongest cost argument against Devin or any metered agent: a team's existing Claude seat becomes the compute layer, and Nyx is the safety and orchestration layer on top.
Competitor pricing gathered via a deep research pass on 2026-05-28 and cited inline in the source markdown at docs/valuation/nyx-market-value-2026-05-28.md. Some URLs come from secondary aggregators rather than vendor canonical pricing pages. Before publishing externally, confirm each figure against the vendor's own page, especially the Devin ACU rate, the Copilot 2026-06-01 usage-based transition, and the Windsurf quota change. No customer testimonials or adoption numbers were invented; where evidence is thin (self-iteration track record, willingness-to-pay) the doc says so.