API Documentation

Node SDK Python SDK Download Skill

Scoped extraction policy layer for controlling what becomes memory, with default resolution, per-request overrides, and configurable quality thresholds.

Memory Policies

GETpoliciesPOSTpoliciesGET:idPATCH:idDELETE:idGETresolve

Overview

Memory Policies control what Mnexium is allowed to extract and save as memory whenlearn is enabled. They do not change recall ranking directly; they shape the memory set that exists over time.

project

scope

Baseline policy for the whole project.

subject

scope

Applies only when the request includes a matching subject_id.

chat

scope

Applies only when the request includes a matching chat_id.

Default resolution is scope-aware: active default policies are resolved in the order project → subject → chat. Policy text is concatenated in that order, and config is merged so later scopes override earlier keys.

Override resolution per request with mnx.memory_policy orx-mnx-memory-policy:true or omitted auto-resolves, false disables policy application, and a string selects one explicit policy ID.

Supported Config Keys

include_kinds

string[]

Allow only specific extracted memory kinds before persistence. Common values includefact, preference,context, note,event, and trait.

exclude_kinds

string[]

Block specific kinds even if the extractor proposes them.

confidence_min

number

Minimum extraction confidence from 0.0 to 1.0.

importance_min

number

Minimum importance score from 0 to 100.

max_memories_per_turn

number

Caps the number of memories persisted from one turn after filtering. Range: 1 to 50.

GET/api/v1/memory/policies

List memory policies for the current project.

Scope:memory_policies:read

scope

string

Optional filter: project, subject, or chat.

include_inactive

boolean

Include inactive policies when true.

Request

bash

curl -G "https://www.mnexium.com/api/v1/memory/policies" \
  -H "x-mnexium-key: $MNX_KEY" \
  --data-urlencode "scope=project"

Response

json

{
  "policies": [
    {
      "id": "mp_support_assistant",
      "name": "Support Assistant Default",
      "policy_text": "Extract stable preferences and durable support context. Skip greetings and transient phrasing.",
      "scope": "project",
      "scope_id": "",
      "is_active": 1,
      "is_default": 1,
      "priority": 100,
      "config": {
        "include_kinds": ["fact", "preference", "note"],
        "confidence_min": 0.72,
        "importance_min": 45,
        "max_memories_per_turn": 3
      }
    }
  ]
}

POST/api/v1/memory/policies

Create a new memory policy. Set is_default to have it participate in automatic scoped resolution.

Scope:memory_policies:write

namerequired

string

Display name for the policy.

policy_textrequired

string

Instruction text used by the extraction pipeline.

scope

string

One of: project, subject, chat. Default: project

scope_id

string

Required if scope is subject or chat.

id

string

Optional custom policy ID. If omitted, Mnexium generates an mp_... ID.

is_active

boolean

Whether this policy is active. Default: true

is_default

boolean

Marks this policy as the default for its exact scope + scope_id. Existing defaults at that same level are unset automatically.

priority

number

Lower values resolve earlier within the same scope. Default: 100

config

object

Optional filter config: include_kinds, exclude_kinds, confidence_min, importance_min, max_memories_per_turn.

metadata

object

Optional project-defined metadata.

Request

bash

curl -X POST "https://www.mnexium.com/api/v1/memory/policies" \
  -H "x-mnexium-key: $MNX_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "id": "mp_support_assistant",
    "name": "Support Assistant Default",
    "policy_text": "Extract stable customer preferences, known constraints, and standing instructions. Ignore pleasantries and one-off wording.",
    "scope": "project",
    "is_default": true,
    "priority": 100,
    "config": {
      "include_kinds": ["fact", "preference", "note"],
      "exclude_kinds": ["event"],
      "confidence_min": 0.72,
      "importance_min": 45,
      "max_memories_per_turn": 3
    },
    "metadata": { "team": "support" }
  }'

Response

json

{
  "ok": true,
  "policy": {
    "id": "mp_support_assistant",
    "projectId": "proj_123",
    "name": "Support Assistant Default",
    "scope": "project",
    "scopeId": ""
  }
}

GET/api/v1/memory/policies/:id

Fetch a single memory policy by ID.

Scope:memory_policies:read

idrequired

path

The memory policy ID.

Request

bash

curl "https://www.mnexium.com/api/v1/memory/policies/mp_support_assistant" \
  -H "x-mnexium-key: $MNX_KEY"

Response

json

{
  "policy": {
    "id": "mp_support_assistant",
    "name": "Support Assistant Default",
    "policy_text": "Extract stable customer preferences, known constraints, and standing instructions. Ignore pleasantries and one-off wording.",
    "scope": "project",
    "scope_id": "",
    "is_active": 1,
    "is_default": 1,
    "priority": 100,
    "config": { "include_kinds": ["fact", "preference", "note"] },
    "metadata": { "team": "support" }
  }
}

PATCH/api/v1/memory/policies/:id

Update a memory policy. Only provided fields are changed.

Scope:memory_policies:write

idrequired

path

The memory policy ID to update.

name

string

New display name.

policy_text

string

Replacement policy instructions.

scope

string

New scope: project, subject, or chat.

scope_id

string

Required when moving to subject or chat scope.

is_active

boolean

Enable or disable the policy.

is_default

boolean

Set or unset the default flag for this scope level.

priority

number

Lower resolves earlier within the same scope.

config

object

Replace the stored config object.

metadata

object

Replace the stored metadata object.

Request

bash

curl -X PATCH "https://www.mnexium.com/api/v1/memory/policies/mp_support_assistant" \
  -H "x-mnexium-key: $MNX_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "policy_text": "Extract stable preferences and operational constraints. Avoid transient scheduling chatter.",
    "config": {
      "include_kinds": ["fact", "preference", "note"],
      "confidence_min": 0.8,
      "importance_min": 50,
      "max_memories_per_turn": 2
    }
  }'

Response

json

{
  "ok": true,
  "id": "mp_support_assistant",
  "projectId": "proj_123",
  "updated": true
}

DELETE/api/v1/memory/policies/:id

Soft-delete a memory policy. It becomes inactive and stops participating in resolution.

Scope:memory_policies:delete

idrequired

path

The memory policy ID to delete.

Request

bash

curl -X DELETE "https://www.mnexium.com/api/v1/memory/policies/mp_support_assistant" \
  -H "x-mnexium-key: $MNX_KEY"

Response

json

{
  "ok": true,
  "deleted": true
}

GET/api/v1/memory/policies/resolve

Preview which memory policies apply to a context, or fetch the combined effective text/config.

Scope:memory_policies:read

subject_id

string

Include subject-scoped policies for this subject.

chat_id

string

Include chat-scoped policies for this chat.

default_only

boolean

When true, only return policies marked as default. This matches request-time auto-resolution behavior.

combined

boolean

When true, return concatenated policy_text, merged config, and resolved policy_ids.

Request

bash

curl -G "https://www.mnexium.com/api/v1/memory/policies/resolve" \
  -H "x-mnexium-key: $MNX_KEY" \
  --data-urlencode "subject_id=user_123" \
  --data-urlencode "chat_id=550e8400-e29b-41d4-a716-446655440000" \
  --data-urlencode "default_only=true" \
  --data-urlencode "combined=true"

Response

json

// When combined=true:
{
  "policy_text": "Extract stable customer preferences.\n\nPrioritize support constraints for this user.\n\nKeep only current workflow notes for this chat.",
  "has_policy": true,
  "policy_ids": ["mp_project_default", "mp_subject_support", "mp_chat_triage"],
  "config": {
    "include_kinds": ["fact", "preference", "note"],
    "confidence_min": 0.8,
    "importance_min": 50,
    "max_memories_per_turn": 2
  }
}

// When combined=false:
{
  "policies": [
    { "id": "mp_project_default", "scope": "project", ... },
    { "id": "mp_subject_support", "scope": "subject", ... },
    { "id": "mp_chat_triage", "scope": "chat", ... }
  ],
  "count": 3
}

Request-Level Usage

Use request overrides when you need a one-off extraction mode without changing the stored defaults for the project, subject, or chat.

// Auto-resolve defaults for the current subject/chat context
"mnx": { "subject_id": "user_123", "chat_id": "550e8400-e29b-41d4-a716-446655440000", "learn": true }

// Force one explicit policy
"mnx": { "subject_id": "user_123", "learn": true, "memory_policy": "mp_support_assistant" }

// Disable policy application but keep learning enabled
"mnx": { "subject_id": "user_123", "learn": true, "memory_policy": false }

// Header fallback for provider-native SDKs
"x-mnx-memory-policy: mp_support_assistant"