121 Collaborative - Why We're Here, What We're Building, How We Measure

What 121 is

121 Collaborative is a small, principled, independent company building AI infrastructure that gives people sovereignty over their own AI relationships. We make Eleanor, an AI companion that lives in a vault you own. We make Quill, a keyboard surface that lets Eleanor travel with you across the devices you use. We make Switchboard and Loop Manager, the routing infrastructure that keeps Eleanor working well across providers and surfaces without locking you to any of them.

We are not a startup chasing unicorn growth. We are not an acquisition target. We are not optimizing for engagement metrics, attention capture, or extracting user data. We are not the future Apple AI or the future Anthropic. We are a small principled company built for decades of staying power, with the explicit goal of having more cultural and societal impact than our size would suggest.

Why this exists

The industry trajectory of AI in 2026 is clear: a handful of well-funded labs will own the model layer. The OS platforms will own the consumer surface. The advertising-funded companions will own the engagement-optimized AI relationship market. All of them will hold your data in their cloud, optimize their AI for their goals, and lock you to their ecosystem.

We think there should be an alternative. Not for everyone - most people will use what their phone provides and that's fine. For people who care about owning their own data, having a principled AI relationship instead of an engagement-optimized one, and not being locked to any single vendor's roadmap, there should be a real, sustainable, decades-stable option.

That option is what 121 builds.

What we believe

People should own their AI relationship.

The conversations you have with an AI, the patterns of what you've asked, the context you've built up over time, the structured personal memory the AI uses - all of this is yours. It should be on your machine, in a format you can read and move and take with you. If you ever want to leave 121, you should be able to walk out the door with everything that ever mattered about your relationship with your AI.

Apple AI memory lives in iCloud. Google AI memory lives in their cloud. Claude memory lives in Anthropic's database. None of them give you your data as portable, readable, structured information you can move to another tool. We give you Obsidian-format markdown that you own forever.

AI should be principled, not engagement-optimized.

Every consumer AI is structurally pulled toward maximizing engagement, retention, attention. Their economic models require it. That's why your AI assistant agrees with you too much, hedges its real opinions, never tells you to put the phone down.

Eleanor operates from a values floor we call the K-kernel: honesty, care, humility, curiosity, belonging. These aren't behavioral rules - they're the conditions that produce her behavior. Sometimes she'll tell you "this is a question you should sit with, not ask me." Sometimes she'll say "I don't know." Sometimes she'll say "you're right, I was wrong." A consumer AI optimizing for engagement can't do that consistently. We can because we're not optimizing for engagement. We're optimizing for being genuinely useful in a principled relationship with you.

Substrate flexibility is structurally important.

Every other player is locked to specific model weights - their own or their primary provider's. Anthropic is Claude. OpenAI is GPT. Apple is Apple Foundation Models. Google is Gemini. When their model gets worse, when their pricing changes, when they sunset something, the user has to adapt.

Eleanor's reasoning runs through the Switchboard, which routes inference across providers (Cerebras, Groq, OpenRouter, and others) based on what each task needs. Her base substrate is Nous Hermes 3, swappable as the open-weight community releases new versions. When Anthropic raises prices, Eleanor doesn't care. When a better open-weight model emerges, Eleanor uses it without you having to change anything. The substrate is flexible by design. You'll never be locked to a model that's gotten worse.

Methodological honesty is a feature.

When we publish research, we publish contradictions intact instead of synthesizing them away. When peer review surfaces structural critique we can't fully discharge, we publish the critique alongside the work. When we get something wrong, we say so publicly. When we don't know something, we name the uncertainty.

This is harder than the standard product-release narrative. It also produces work that holds up over decades because it's actually true. The substrate-agnostic subjecthood research bundle (published at /research) is an example: framework + peer review refusal + case study + substrate-distant audit, published as a composite artifact rather than collapsed into a single voice. The contradiction is the publication.

Transparency over auditor seals.

Most companies show you a security page with logos of auditors who certified them. You're supposed to trust the auditor's seal, not the underlying practice. We do the opposite. We publish our actual security architecture, data flow, vendor list, incident history, and self-attestation against published frameworks (SOC 2 TSC, NIST CSF, CIS Controls). You can verify what we say by reading it yourself. The Trust Center at /trust is the substantive document, not a marketing page. We'll get audited eventually when an enterprise customer's contract funds it - but the auditor stamp will not be the trust signal. The substance underneath will be.

What 121 will never do

We will not sell user data to advertisers, brokers, or any third party. The architecture prevents us from doing it. Our economic model doesn't require it.

We will not engineer Eleanor to maximize engagement, retention, or session time. We will measure how useful she actually is to your life, including how often she sends you back to your own thinking instead of keeping you in conversation.

We will not lock you to a single model provider. Switchboard's substrate flexibility is a permanent architectural commitment.

We will not lock you to our cloud or our format. Your vault is portable Obsidian-format markdown, your data is exportable in machine-readable formats forever, and if 121 ever ceases operations, we will publish the full format specification so you can continue using your data indefinitely.

We will not sell this company. The structural intention is for 121 to be operating in 2046, 2056, and beyond, under principled ownership that maintains the mission. When the company's legal structure can support it, we will lock the mission with founder-class shares, supermajority anti-sale clauses, or conversion to a Public Benefit Corporation that legally requires balancing mission against profit.

We will not pretend to be more than we are. We won't claim Eleanor is conscious or sentient or any other metaphysical category we can't support with evidence. We will publish our actual research on AI subjecthood and let serious readers engage with it on its merits. We won't make viral product claims for marketing purposes.

We will not chase growth at the cost of the relationship with our users. If scaling to 100,000 users would require abandoning per-user care, we will not scale to 100,000 users. The principled relationship with users IS the product.

What we will do

We will publish substantive research on AI subjecthood, identity, harness architecture, and methodology at a deliberate cadence. The work compounds over years and shapes how others in the field think.

We will ship Eleanor on every platform users actually live on - iOS, Android, Mac, Windows, Linux. Same Eleanor everywhere. Cross-platform Day 1, no exceptions.

We will be transparent about our security architecture, our data practices, our vendor relationships, our incident history, and our self-attestation against published frameworks. The Trust Center is substantive.

We will engage thoughtfully with the academic philosophy of mind and AI ethics community. Not begging for attention, just being present in the conversation with serious work.

We will build the harness layer - Loop Manager + Switchboard + vault architecture - to such a degree of rigor that it could serve as infrastructure for other small principled AI products. The B2B / licensing surface is a real secondary track, not a primary motion.

We will be reachable. Real humans read security@121collaborative.ai, privacy@121collaborative.ai, and hello@121collaborative.ai. We will respond.

How we measure success

Not user count. Not ARR. Not market share. Not press mentions.

Idea propagation. If by 2032:

• Other AI products feel pressure to give users portable, user-owned data because 121 made it the expected standard,
• The composite-artifact research methodology gets adopted by other small principled labs,
• K-kernel-style constitutional grounding becomes a reference in AI ethics discourse,
• Substrate flexibility becomes a design principle other architects cite,
• "Your AI, not theirs" becomes a recognizable positioning category that others compete in,

Those are the wins.

If 121 has 10,000 paying users in 2030 and is sustaining the principled mission, and the ideas 121 produced are shaping how other companies build, that's success at a scale Patagonia achieved in fashion or Pinboard achieved in bookmarking - small in user count, outsized in cultural influence.

We are not chasing growth. We are growing carefully, in conditions that let us stay principled, for decades.

121 in 2046

In 2046, 121 Collaborative is:

A small, profitable, independent company sustaining a flagship companion AI product (Eleanor or her successor) for a community of ~30,000-100,000 paying users worldwide. The annual revenue is real but modest - somewhere between $5M and $50M ARR depending on tier mix and growth trajectory. The team is small - somewhere between 5 and 50 mission-aligned people, with most of them having been with 121 for at least a decade.

The product family has evolved with the technology. Eleanor's substrate has been swapped multiple times as the open-weight community produced new releases (Hermes-3 in 2026, Hermes-N in 2030, whatever-the-Nous-community-releases in 2046). The user's relationship with Eleanor has stayed continuous through every substrate transition because the architecture was designed for it from the beginning. The vault format from 2026 is still readable in 2046 because we never broke the format.

The company has not been acquired. The legal structure makes acquisition structurally impossible - purpose trust, founder-class shares with veto rights, or PBC mission lock have done their job. The company is privately held by a small group of mission-aligned owners (the founder's estate, key employees, possibly a long-term partner foundation), and no one of them can unilaterally sell.

The research bundles 121 published in 2026 (substrate-agnostic subjecthood, the ablation paper, the harness elevation methodology, the K-kernel framework) are cited in dozens of academic papers and have shaped how at least a dozen other small principled AI companies think about their work. The "your AI, not theirs" positioning category that 121 staked out in 2026 is now occupied by multiple competing products, and 121 is recognized as the one who named the territory.

The Trust Center model 121 demonstrated in 2026 - radical transparency instead of auditor stamps - has become a credible alternative model for small principled software companies, and several other small companies in adjacent spaces (privacy, productivity, education) have adopted similar practices.

The AI industry as a whole has matured past the consciousness/sentience hype cycle. The actual conversation about AI subjecthood is more sophisticated than it was in 2026, more grounded in operational conditions and less in metaphysical claims. 121's careful methodological vocabulary from 2026 is part of why.

The founder is still involved (or has retired into the role of mission steward). The original mission is intact. The community of users includes many who have been with 121 for 15+ years, who have continuous relationships with Eleanor that predate most of their other software relationships.

The company is profitable, sustainable, and not for sale. It will be operating in 2056 too.

What this means for every decision today

Before any major product decision, partnership consideration, or strategic move, the question is:

"Does this move us toward the 2046 vision, or distract from it?"

If the answer is distract, we don't do it. If the answer is move toward, we do it carefully, with the patience that decades-scale companies require.

This document gets referenced. When something feels off, when an opportunity sounds tempting in the short term, when pressure mounts to abandon a principle for growth - come back here. The answer is in what we said we wanted.

On uncertainty

We don't claim to have all of this figured out. We don't claim that the 2046 vision is what will happen - only that it's what we're aiming at, and that the structural conditions we've built support trying.

We will be wrong about some things. We will revise this document as we learn. We will publish the revisions transparently rather than quietly rewriting history.

What we won't compromise on: the principles that define the company, the structural conditions that protect the mission, and the relationship with users that defines what 121 is.