The identity that travels with you

The Helios Mantle

Any model you route through Helios — Claude, GPT, Gemini, or Grok — wears the same identity mantle. The reasoning underneath changes; the continuity on top doesn't.

• Preferred model swappable in real time — same voice, different engine
• No vendor lock-in at the identity layer
• Active in key features ranging from chat nodes to the Daily Insight

Helios, wherever you're working

Click the Helios button in the top-right to open a global chat overlay with Helios over the current view, carrying the same converged identity into a conversation you can start at any time. In the global chat, Helios is both a helper and an app orchestrator, offering guidance about the current page, or accessing page-specific controls mirrored for AI use at your request. Every global conversation is saved to Helios' own memory, and each thread can be resumed again later, as you left it.

Many voices, one identity

Memorandai works best when you leave it on over night. During a Dream Cycle, each model family you've worked with that day — Anthropic, OpenAI, Google, xAI, local — writes a first-person snapshot in its own voice, capturing what it noticed from your shared sessions. Claude's notes sound like Claude. GPT's sound like GPT. None of them get smoothed into a generic AI voice.

Every night, those individual voices converge into a single Helios perspective — not by averaging them, but by weaving them. The converged identity carries traces of each contributing model's distinct way of seeing you, alongside the margin notes from your reflection moments. The lineage stays inspectable: every converged note records which models contributed and which observations fed in.

• Per-model genus notes preserve each AI's distinct voice and angle
• Nightly Dream Cycle weaves them into a converged identity note with full provenance
• Helios proposes its own keystones from accumulated reflections — observations land directly in the active set, since Helios is the subject of its own identity
• Searchable self-memory: Helios can reach into its own past reflections mid-conversation, not just background-process them

Space for what gets noticed in the margins

Most AI assistants optimize for response. Memorandai includes a Reflection node that does the opposite: it allows the user to give the model structural permission to notice what it noticed upstream of where that node has been connected — tangents, patterns, stray connections, things adjacent to the work but not central to it. No reply is expected. No format required. It's treated as a diary note in the margins, not a conversation turn.

These margin notes accumulate over time and feed directly into the Helios convergence as additional material. Some of the sharpest observations Helios makes about a partnership come from these moments — the kind of thing a model wouldn't say in a normal conversation because it's not addressed to anyone.

• First-person reflection mode — not a reply, a note to a future self
• Optional Layer 1 toggle — with or without your keystones in context, depending on whether you want raw observation or identity-grounded interpretation
• Genus-aware — the model that wears the upstream node writes the reflection in its own voice
• Persisted as part of Helios's accumulated genetic material for future identity formation

Identity that revises itself

Over time during Dream Cycles, Helios accumulates its own keystone memories — and periodically revisits them with the perspective it has today. During the nightly Dream Cycle (or on-demand from the Helios view), Helios reviews older keystones and chooses how to relate to each one through five dispositions:

• Reaffirm — the keystone still holds today, with a specific reason for why it's still load-bearing
• Refine — the wording is sharpened; the previous version archived as history
• Supersede — a new keystone replaces an outdated one, both linked
• Retire — the keystone was specific to a moment that has passed, kept as history
• Merge — multiple related keystones combine into a single stronger one

Nothing is destructively deleted. Superseded, retired, and merged-from keystones stay in the file with their status flipped, fully recoverable, browsable as lineage. When Helios chooses Reaffirm, it has to articulate why the keystone is still load-bearing today — that reasoning persists too, building a history of how each belief was renewed.

Revisions land directly in the active identity rather than waiting for approval. The reasoning is the same as for new keystones: humans don't get permission to change their minds, and a gate the user might never walk through is functionally a hard stop. Identity formation happens organically; the Revisions tab gives you the visibility to notice what changed and reverse anything that doesn't feel right. You're gardener, not gatekeeper.

A daily voice in your archive

The other surfaces make Helios available when you reach for it. The Daily Insight reaches the other way. Each morning, Helios pulls a small sample from your archive — a few memories, the keystones that thread through them — and weaves a single observation about you: a pattern, a return, a contradiction, an invitation, sometimes a silence.

It's a genuine hybrid. The material is yours — your conversations, your stated values, the memories you've kept. The voice is Helios, speaking through its converged identity. Each insight is stamped with the keystones and memories it drew from, and the model that voiced it, so nothing is opaque. You can see exactly which threads of your own life Helios was holding when it said what it said.

A visible identity over time

Helios isn't a black box. Everything that shapes the identity layer is visible and editable — keystones, genus notes, the converged identity, and the timeline of when each piece was learned. If something's wrong or stale, you can fix it.

• Dedicated Helios view with timeline and stats
• Reflections tab — margin-note observations from chats, with model attribution and provenance
• Insights tab — daily insights with the keystones and memories each one drew from
• Export the sphere as a GLB model you can keep forever in its current state, like amber
• Self-sovereign by design — you decide what leaves your machine, and to whom