One developer surface for every Neuraphic model.

Open Console, create a project, generate one API key. The same key calls Praeth for continuous risk scoring, Claeth for input integrity, and Prion for autonomous defense — full size or Prion mini. One bill, one audit log, one quota dashboard.

Every new project starts with $100 in credits — no card required to begin. Failed requests are never billed. Every model returns within sub-second on the median, and every one of them was trained on traffic from Neuraphic's own production surfaces before it ever reached you.

Models

The Neuraphic model line.

Every model is reachable from a single project under a single API key. Pricing is per token — input and output billed separately, so a long prompt with a one-word reply doesn't pay for tokens that never came back. Failed requests are never billed. Versioned cleanly: pin praeth-1.0-001 or float on praeth-1.0 for production.

Pricing

Start with $100. Stay free until you scale.

Every new project gets $100 in credits at signup. No credit card required to start. The credits work across every Neuraphic model — burn them on Praeth, save them for Workers, mix and match. Whatever's left after thirty days does not expire.

Pricing is per-call, not per-seat. Failed requests are never billed. Streaming endpoints bill the same as their non-streaming counterparts. Live and test keys share the same SDK surface, so promoting from prototype to production is a one-character change.

  • $100 in credits on every new project — no card required
  • Per-token pricing — input and output billed separately
  • Failed requests are not billed
  • Test keys are free at any volume — for evaluation only
  • Quota auto-bumps on usage signals (opt-in)
  • Enterprise tier ships with private deployment + audit-log export
ModelInputOutput
Praeth$0.50$1.50
Claeth$1.00$2.00
Prion mini$1.50$5.00
Prion$5.00$25.00
All prices per 1M tokens. Failed requests not billed.
Approach

We're building an AI ecosystem — from the models to the infrastructure to the products that run on them.

Neuraphic runs a single stack. The agents that do the work, the defenses that inspect every input, and the infrastructure beneath all of it are built inside one organization. Research and product sit at the same table. Nothing gets bolted on the week before launch. From the first commit, every system is written to be read, supervised, and trusted by the time it reaches a customer.

Trust & safety

Built to be read, supervised, and trusted.

Hash-chained audit log

Every state-changing operation — project creation, key generation, role grant, quota bump, model registration, abuse action — is appended to a tamper-evident log. Quarterly export to immutable storage. Verifier ships in the SDK so customers can replay our chain on their own infrastructure.

Anti-enumeration by default

Internal endpoints return 404 on auth fail, not 401 or 403, so an attacker cannot enumerate which resources exist. Constant-time bearer comparison. Cross-tenant predicates baked into every repository signature. The threat model is published and updated quarterly.

Failed requests not billed

If we 4xx or 5xx your call, you don't pay for it. Quota is enforced before the model backend sees the request, and metering only fires after a successful response. The bill matches the value delivered — never the noise.

Per-product blast radius

Praeth, Claeth, Prion, and Workers each run in their own AWS account with their own VPC, KMS keys, and IAM boundary. A compromise in one cannot reach the others. Console is the only surface that crosses boundaries — and that boundary is itself audited continuously.

SOC 2 Type II in flight

Audit window opened Q2 2026. Targeting Type I report Q3 2026 and Type II twelve months later. Existing controls are documented in the security policy library and reviewed before every release. SOC 2 will cover Console, the four production model surfaces, and the supporting AWS infrastructure.

Responsible disclosure

Encrypted reports to security@neuraphic.com via the Curve25519 GPG key published at /.well-known/security-pgp.asc. Acknowledgment within 24 hours, fix or workaround within 7 days for high-severity issues. Coordinated public disclosure on a 90-day default timeline.

Resources

Built for the people who ship code.

Get started

Open a project. Send your first scoring request before the next standup.