Skip to main content

​
Request

GET https://api.spectralmemory.com/decode?user_id=alice&label=USER.gpu
Authorization: Bearer YOUR_API_KEY

​
Query parameters

ParameterTypeRequiredDescription
user_idstring✅Your user identifier
labelstring✅Channel label to decode
use_modelboolean—If true, decode via model inference instead of state. Default: false

​
Response

{
  "user_id": "alice",
  "label": "USER.gpu",
  "value": "RTX5090",
  "source": "state",
  "latency_ms": 0
}
FieldDescription
valueDecoded fact value
source"state" (fast path) or "model" (inference path)
latency_msInference latency. 0 for state path. ~400ms for model path.

​
Fast path vs. slow path

Fast path (default, use_model=false): reads the value directly from persistent server-side state. Instant. Use for routine agent queries. Slow path (use_model=true): sends the .mem block to the Hermes3 inference endpoint. The model performs implicit frequency decomposition to recover the value. Use to verify spectral encoding integrity.

​
Errors

CodeReason
400Unknown label
401Invalid or missing API key
404No FDM block found for user_id (slow path only)

​
Notes

  • Does not count against your monthly encode limit.
  • The slow path falls back to DSP-based decoding if the Modal inference endpoint is unavailable.