Memory for agents.
Control for you.

Create a Nucleus Read the quickstart

Scoped agent memory, bounded context and reviewable decisions. Your models. Your boundaries.

Agent memory

Carry useful knowledge into the next task. Isolated Nuclei and permissioned Scopes keep recall inside your boundaries.

ScopeMemory Cell

engineeringKeep changes focused

engineeringReview before rollout

Memory Cells · Scoped recall · Tombstones

Context Capsules

Give each model call relevant summaries within an explicit token budget.

Bounded context

Temporal knowledge

Keep relationships, provenance and contradictions visible as knowledge changes.

Time · Evidence · Confidence

Consequence Thought

Compare bounded scenarios with assumptions and uncertainty available for review.

Decision support, with human approval

Outcome learning

Separate predictions from observations. Review sensitive lessons before memory write-back.

Observe · Reflect · Review

Governance

Control access and deletion. Require human approval for high-risk actions.

Permissions · Audit · Review

REST, MCP & SDKs

Connect the stack you use. Your application keeps provider credentials and final model calls.

Model-independent integration

Start with memory. Connect context, knowledge and decisions as your agent needs them.

Built around your workflowREST MCP TypeScript / JavaScript Python PHP Go Rust CLI

Explore the platform

See what your agent
carries forward.

Search scoped memory, assemble a bounded context pack and compare reviewable decisions. Try the local examples below.

Example workspace / AtlasNeutron AI

Nucleus / atlas

A little memory. A better next step.

Explore a few example Memory Cells. Recall stays inside the Scope you choose.

Preference

Keep changes focused

Use small modules and test behaviour before a release.

engineering

Decision

Review before rollout

Require a human review before changing production access.

engineering

Lesson

Check the live result

Verify the deployed version and its public routes after release.

engineering

Knowledge

Keep context bounded

Return relevant summaries within the requested token budget.

engineering

4 example summaries · Deleted examples stay out of recall

Illustrative data · Runs locally · No account or API token needed

Explore agent memory

Your stack. Your model.

Useful context,
one API call away.

Request a scoped context pack from your application. Keep model credentials and final model calls with your existing provider.

Use Neutron AI with
TypeScript / JavaScript.

1Create a Nucleus.
2Remember inside a Scope.
3Request bounded context.

Open the quickstart

Authenticated REST examples. Create the example Nucleus and Scope first. Keep API tokens on your server.

context.mjs

const response = await fetch(
  'https://neutronai.dev/api/v1/memory/agent-context',
  {
    method: 'POST',
    headers: {
      Authorization: `Bearer ${process.env.NEUTRON_API_TOKEN}`,
      'Content-Type': 'application/json'
    },
    body: JSON.stringify({
      "nucleusId": "atlas",
      "scopeIds": [
        "engineering"
      ],
      "agentId": "release-assistant",
      "task": "Prepare the next release",
      "tokenBudget": 1200
    })
  }
);
if (!response.ok) throw new Error('Context request failed');
const context = await response.json();

POST /v1/memory/agent-contextRead the API docs

SDKs are release candidates and typed coverage varies. Explore SDK, CLI and MCP workflows; the examples above use the authenticated REST API.

Under the surface

Continuity needs
a dependable foundation.

Keep hot state authoritative and background work explicit. Deletion writes tombstones so known retrieval and asynchronous paths cannot restore forgotten memory.