Compression

Neutron AI is designed to turn large interaction histories into small, task-ready context. Compression keeps agent prompts focused, reduces repeated tokens, and makes long-running integrations easier to maintain.

What Gets Compressed

Neutron can compress:

• conversation outcomes
• user and workspace preferences
• agent working notes
• repeated support or sales context
• project summaries
• knowledge snippets
• tool results that are useful later

Public Behavior

When your app stores memory, Neutron AI can:

• summarize verbose text into reusable memory
• preserve source labels and scope metadata
• reduce repeated context before future tasks
• rank the most relevant memory for a task
• omit memory that is stale, deleted, or outside the requested scope
• return warnings when a context pack should be handled carefully

Compression Is Not Authorization

Compression makes memory smaller. It does not replace your product’s authorization checks. Your app should decide which tenant and scope IDs the current user or agent may use before calling Neutron.

Usage Pattern

await client.remember({
  scopeId: process.env.NEUTRON_SCOPE_ID!,
  type: "summary",
  text: "The onboarding team prefers short implementation plans with clear owner and date fields."
});

const context_pack = await client.agentContext({
  scopeIds: [process.env.NEUTRON_SCOPE_ID!],
  agentId: "agent_onboarding",
  task: "Draft the next customer onboarding checklist."
});

Use context packs as supporting memory, not as the only source of truth. Current user input, live records, and retrieved evidence should win when they conflict.