Northsun · voice

Northsun is the memory and enrichment layer for AI agents. Its first app is voice.

One call before your agent speaks returns the caller profile tied to the number and what happened on recent calls.


How a call works

  1. 01
    The phone rings
    +1 415 555 0147 (Example number)
  2. 02
    One call out
    POST /v1/recall_and_enrich
  3. 03
    What comes back
    caller · memory · enrichment
  4. 04
    Then the agent speaks
    with the context that came back

Recall before the turn

One call before your agent speaks returns the caller’s profile and recent calls, when Northsun has them. It is designed for sub-300ms voice recall.

Write back after hangup

After hangup, one call writes the conversation back. Northsun answers 204 when it accepts the write.

Unknown callers are not errors

A number Northsun has not seen before comes back as known: false, not as an error. No guess is made.


The control plane

See the docs →

Memory governance

Design stage, not shipped

Governed facts will be versioned. When a fact changes, Northsun will store the new version beside the old, and earlier versions will stay readable as of their date.

Response details

Each successful recall response includes a trace ID and timing_ms. When memory comes back, a freshness value says how long the answer had been cached.


Deterministic by choice

Constrain the model and you get a script. Constrain the memory and you keep the agent.

Northsun Voice

Where it fits

Read the docs →

Your voice stack makes two HTTPS calls: one before the agent speaks, one after hangup.


Receipts, not benchmarks

We don’t publish a benchmark. Each successful recall response carries its own timing_ms.


Early access. We’re onboarding design partners by invitation.

Get access

Choose Northsun as your agent memory layer.