Singularity Gate

Type a structured task. The first matching call compiles a Wasm skill. A later call with different numbers uses that skill. The raw POST /v1/execute route is still there.

Embedder trigram-hash-v1 Fixed model 11.44 USD/month Tier 0 requests 2 Gossip peers 0, propagation 0 ms DAG branches 1 Prefix submit after a 0.84 hit: 0 us

This process

Measured on the gateway that answered. Tier 2 is the compile. Tier 0 is the replay.

Waiting

Comparison price

Not a live provider call. saved_usd adds 0.002 USD per tier 0 hit unless OPENAI_COMPARE_USD_PER_CALL is set. That number is not an invoice. Provider latency was not measured, so this panel does not tick a fake timer.

$0

Ledger and snippets

Signing up...


Commons

The commons record drops the numeric payload, the raw text, and the tenant id. It is not a zero-knowledge proof. The DAG on this node still stores the task string and the stdout.

No commons records yet.

Proof and Wasm

Run a skill to load its commons record. Lean checking a Nat or Bool file does not certify the Wasm bytes. connects_lean_to_wasm stays false.

Catalog

Upgrader stages

  1. Intercepting: tier 1, no compiled skill, or the circuit breaker.
  2. Shadow crystallizing: tier 2, first compile. Catalog samples must match before the module is stored.
  3. Equivalence: those sample vectors, not a hidden provider transcript.
  4. Crystallized: tier 0 replays. Changed numbers keep the same structural signature.

No run yet.

Plans

Starter, Pro, and Enterprise are labels on the local key. Card checkout is blocked until Adolfo sets STRIPE_SECRET_KEY and STRIPE_WEBHOOK_SECRET.

Five uses, and what this build actually does

1. Endpoint upgrade: chat completions compile a matching skill and later calls skip the model.

2. Guardrail gossip: a local three-process gossip run moved one payload. It was not a planetary push.

3. Edge export: download module.wasm or the .slx bundle. Browser exec time is whatever performance.now reports on that device.

4. Audit: the response includes the Lean check flag and connects_lean_to_wasm false.

5. Royalties: not paid out. saved_usd is the comparison constant times tier 0 hits.

Bill arithmetic

Recent requests

TierTaskLatency usEst. tokensModel cost USDt ms
0831acead3e952454e0f8baf2defd18a65702317f4d8ce0534251723b335d806526906.277e-101794671109954
0831acead3e952454e0f8baf2defd18a65702317f4d8ce0534251723b335d80653745508.739e-81794671109679
Advanced: raw JSON to /v1/execute

  

Prediction hits 0, misses 0. Submit without a prefix step: 0 us. RLVR before 0.000, after 0.000. Requests 2. Estimated tokens 0. Model calls 0. Cap used by the cost gate: 12 USD. BitNet on this tree: local generation 20.7 tokens/s at 1229.6 MB, live 16.85 tokens/s on 2 threads. TL2 is off because the GGUF is I2_S. Neither run met 35 tokens/s.