1. API
2. Docs
3. Memory scopes

Memory scopes

A memory scope lets MUGEN remember your project between requests, in synapses rather than resent context.

Using a scope

Add "memory_scope": "your-scope" to a chat completions, responses, messages or /v5 turn request. Within the scope, MUGEN remembers what was asked and what it was about, and can refer back to it. Without a scope, a request contributes only to general capability.

A scope name is up to 128 letters, digits, ., _, - or :. Scopes belong to the key that created them; another key cannot see them.

Managing scopes

The native API treats scopes as resources: list them with GET /v5/memory-scopes, create one with POST /v5/memory-scopes, inspect one with GET /v5/memory-scopes/{name}, and forget one with DELETE /v5/memory-scopes/{name}. Forgetting is final.

Isolation

Synapses that encode organisation-specific knowledge are partitioned by tenant; general capability is shared. Isolation between scopes is verified continuously rather than proved. Don't place anything in a scope that you would not accept MUGEN learning from, and use one scope per project: MUGEN will draw connections within a scope that you may not want across projects.