CRESCENT
Open intelligence workspaceNo token required

Your mind. A wider orbit.

An open intelligence workspace. Think with Nemotron, build with agents, and keep a record of every run.

Nemotron route · …  ·  Receipts · unsigned

Surfaces

Six instruments. One record.

01

Chat

Streamed conversations with Nemotron. History stays in your browser. Every reply names its provider.

02

Agents

Character profiles with their own instructions, allowed routes and opt-in local memory.

03

Orbit

Plan → execute → review missions under step and token caps. Cancel any time.

04

Foundry

Turn good work into versioned skills. Promote only after deterministic checks pass.

05

Proof

Signed receipts that hash both request and response. Verify in your browser.

06

Studio

Upload an image, get a close reading, a brief, or a prompt draft from Super VL.

Orbit · mission control

One task.
The right orbit.

Orbit plans a mission, runs each step on the route you allow, then asks a reviewer pass to check the result. You set the step and token caps. A review is a check — not proof of correctness.

Open Orbit
missionBUDGET · STEPS · ROUTEPLANroute-checkedEXECUTEcapped stepsREVIEWa check, not proof

Proof

A receipt for every run.

Crescent signs a statement of what it asked, which model answered and how long it took — with SHA-256 commitments to both the request and the full response text. A signature authenticates Crescent's statement. It does not prove the answer is correct, which GPU ran it, or that the provider retained nothing.

Verify a receipt
Example · illustrative, not a real runcrescent.receipt/1
{
  "requested_model":  "nvidia/nemotron-3.5-lightning-30b-a3b",
  "actual_provider":  "nvidia-hosted",
  "route_reason":     "auto: short text → Lightning",
  "status":           "completed",
  "usage_source":     "upstream",
  "cost":             null,   // "not priced"
  "request_sha256":   "9f2c…e41a",
  "response_sha256":  "4b7d…02c9",
  "attestation":      { "status": "none" },
  "signer":           { "alg": "Ed25519", "origin": "server" }
}

Models

Named routes. No quiet swaps.

NVIDIA hosted API

Nemotron 3.5 Super VL

Deep reasoning · tool use · vision

Context · 262K (hosted endpoint)

NVIDIA hosted API

Nemotron 3.5 Lightning

Fast text tasks

Context · See NVIDIA model page

Lovable AI Gateway

GPT-6 Astra (Preview AI)

Opt-in preview route — not Nemotron

Context · Provider-managed

Self-hosted runtime

Bolt 74B-A9B

Local / private deployments

Context · Depends on your runtime

External Hermes API server

Hermes Agent runtime

Real Hermes harness · skills · memory · tools

Context · Set by the Hermes inference provider

Full catalog & status

Privacy, plainly

Your history stays on your device.

  • Local by default

    Chats, agents, skills and receipts live in this browser's storage. Crescent's server stores no transcripts.

  • Providers do process prompts

    When you send, the model provider you chose processes your prompt under its own terms.

  • Ephemeral chats

    Mark a chat ephemeral and it is never written to storage.

  • No confidential badge

    This deployment performs no hardware attestation, so it makes no confidential-compute claims.