Add governed memory to your application

Use DoubleGate to submit observations, inspect their review state and retrieve knowledge with provenance. The SDK gives Python applications a client; agent hosts can connect directly through MCP.

Choose your starting point

What you are building	Start here
A Python application using memory	Connect, propose and recall
An agent using tools	Connect your agent through MCP
A deterministic content check	Author a check
A DoubleGate integration or extension	API reference

One client, a few memory operations

import os
from doublegate_sdk import connect

client = connect(os.environ['DOUBLEGATE_ENDPOINT'],
                 token=os.environ.get('DOUBLEGATE_TOKEN'))
answers = client.recall('How do we label laboratory specimens?', limit=5)

Start with a running gate endpoint. Enable writes explicitly when your application needs to propose observations. Preserve the artifact id to follow review, and keep provenance with retrieved content. See the complete quickstart.

How the workflow fits together

Submit an observation → inspect its state → retrieve it when eligible for serving.

The gate performs review and enforces permissions. Your application chooses when to submit and how to use the returned evidence. Pending submissions do not become knowledge merely because a request succeeded.

Developer resources

The development SDK is distributed through GitHub. Remote authenticated MCP and release acceptance are tracked in the roadmap. The inspected local Client Gate's MCP listener is keyless on loopback; do not expose it as an authenticated remote API.