SupraOS — How It Works

OVERVIEW

Four pillars. One system.

Every AI product on the market is a piece of a team. A chat window is a conversation. A browsing tool is an assistant. A coding agent is a developer. SupraOS is built the other way around. We started from what a real company actually is, then broke it into the four things a team needs to function: a shared memory so work carries forward, specialized roles so the right agent does the right job, visibility into what's happening so decisions are informed, and a way to turn conversations into shipped work.

Memory gives the team continuity. Without it, every conversation starts from zero and every lesson is forgotten by Tuesday. Agents give the team specialization. Without it, one generalist tries to do every job and does each of them badly. Data gives the team visibility. Without it, agents make confident decisions on stale assumptions. Build gives the team execution. Without it, you have a team that can plan but cannot ship.

The pillars are not features sitting next to each other. They are one system. Memory feeds the agents. Agents read and write the data. Data frames what the build phase produces. Each shipped piece of work becomes a new memory, and the loop closes. Pull one pillar out and the team stops behaving like a team. Keep all four wired together and the agents start behaving like people who have worked together for years.

Each pillar is useful on its own. Together, they are what makes an AI team feel like a team.

01

Memory

Continuity across every session

02

Agents

Roles that do the work

03

Data

What the team sees

04

Build

How work ships

Memory

A team that actually remembers.

Because a team without continuous memory isn't a team. It's a rotation of strangers.

Every AI tool on the market forgets you by Tuesday. Every chat starts from zero. Every decision your agent made last week has already been erased. A real team doesn't work that way. A real team remembers what was decided, why it was decided, what worked, and what didn't, and it carries that forward into the next conversation, the next department, the next quarter. Memory is the difference between twenty-two strangers and a company that knows itself. Here is how SupraOS gives your team the memory a real organization runs on.

Why memory matters

Without continuous memory, every agent resets at the end of every session. Context evaporates. Last week's decisions have to be re-explained. Mistakes that were already made once get made again. You become the memory of the company, which means you become the bottleneck of the company.

How SupraOS remembers

Every observation, decision, and outcome gets written into a shared memory store and ranked by a seven-factor model called MemoryRank. Access frequency captures what the team keeps reaching for. Outcome correlation captures which memories were associated with good results. Recency, usefulness, and cross-references round out the day-to-day relevance. Consensus tracks when multiple agents agree on the same observation. A pin bonus protects the things you've explicitly told the team never to forget.

When an agent starts a task, memory surfaces through a five-phase Assembly Pipeline. The system gathers the candidate memories, ranks them against the current task, weights the most relevant ones, compresses them to fit the model's context, and hands the finished brief to the agent. The agent starts every job already caught up. Red-line directives, the things you wrote with words like never and must not, bypass the context budget and show up every single time.

The Hindsight Loop closes the circle. After a task ships, outcomes get tagged back to the memories that informed it. Memories that led to good results get reinforced. Memories that led to bad ones get weakened or flagged as Lessons Learned, which become permanent records the team cannot re-repeat.

While the system is idle, the Dreaming Cycle consolidates the week's experiences into higher-order insights, the same way sleep consolidates human memory. Your team wakes up smarter than it went to bed.

MemoryRank

Not all memories are equal. Seven weighted factors determine which ones surface first — so the agent's context window is always filled with what actually matters.

25%Access frequency

20%Outcome correlation

15%Recency

15%Usefulness self-eval

10%Cross-references

10%Consensus

5%Pin bonus

MEMORY LIFECYCLE

REINFORCED

Used in a task that shipped well. Rank goes up.

DECAYED

Not accessed in weeks. Quietly fades.

SURFACED

Similar task starts. Pattern is found and surfaced.

LESSON LEARNED · NEVER DECAYS

Don't publish to the blog before Co-CFO runs legal-scan on claims. Missed once in March; cost a correction post.

SOURCE: Co-CMO · CONSENSUS: 3 AGENTS

Agents

A full organization on day one.

Twenty-two agents, seven departments, and a Co-C-suite that runs the business with you.

A company isn't a pile of identical workers. It's a team with roles, ranks, and real authority. Your Co-CFO doesn't write your blog. Your Co-CMO doesn't chase yield. Your Sentinel isn't your Planner. SupraOS ships with twenty-two agents across seven departments, six core agents that run the platform itself (Guide, Sentinel, Technician, Planner, Recon, Support), and a Co-C-suite that runs the business (Co-CRO, Co-CTO, Co-CMO, Co-CFO, Co-COO, Co-CBO). Every role has a mandate. Every role has a voice. Every role can act. Meet the team.

Why roles matter

A pile of identical generalist agents is not a team. It's twenty-two chat windows with the same personality, each waiting for you to dispatch it. Real teams work because people have specialized mandates, owned areas, and the authority to act inside them without checking with the CEO first.

The 22 agents

Six core platform agents run SupraOS itself. Guide onboards you and progressively surfaces features you haven't tried yet. Sentinel monitors every request and blocks threats before they touch your data. Technician builds custom pages and modifies the OS through conversation. Planner designs workflows from plain-English descriptions and knows every one of the seventy-two node types. Recon browses the web, scrapes sources, and produces structured research briefs. Support resolves tickets end-to-end and escalates only what actually needs you.

The Co-C-suite runs the business. Co-CRO owns research and competitive intelligence. Co-CTO owns engineering, deploys, and the boring mechanics of shipping product. Co-CMO owns marketing, content, and voice. Co-CFO owns finance, runway, and, for operators who need it, portfolio and yield. Co-COO owns operations, vendors, logistics, and calendar. Co-CBO owns business development and the outreach layer.

Agents hand off to each other with full context briefings. When Recon finishes a competitive sweep and passes it to Co-CRO, the handoff includes the watchlist, the anomalies, the confidence level on each, and the relevant memories. When Co-CRO sends a draft to Co-CMO to write up, the brief carries your voice, your audience, and your past takes. No agent starts cold.

The Agent Journal is the social learning layer. After a task, each agent writes what it did and what it learned. Other agents read the journal, acknowledge, build on it, or disagree. Patterns that hold across agents become shared knowledge. The team gets smarter as a team, not just as individuals.

Reach your team anywhere

DM your Co-CFO, brief your Co-CMO, ping the whole org — every department is a message away. SupraOS speaks the apps you already live in.

Discord

Slack

+ voice · SMS · email — same agents, every channel

AGENT HANDOFF · TASK BRIEFING

RECON

Scans

847 sources, 4 anomalies found

CO-CRO

Synthesizes

Ranked thesis on anomaly #2

CO-CMO

Writes

Long-form post in your voice

BRIEF: anomaly #2 (protocol X fee model)

CONFIDENCE: 0.78 · corroborated across 3 sources

MEMORY: pattern matched March '26 observation

VOICE: your back-catalog, newsletter register

DEADLINE: ship by Friday

Data

What your team sees. And what nobody else does.

A team that can't see the state of the business makes bad decisions. Give yours the view.

A real team runs on what it can see. Positions, P&L, risk, knowledge feeds, competitor moves, health metrics, whatever your business actually depends on. SupraOS gives every agent the data it needs, routed through your keys, encrypted in your memory, and surfaced in dashboards only you and your team can see. Swap the underlying model whenever you want: Llama one day, Claude the next, Ollama on your own hardware. Your data stays yours. Your organization stays intact. The brain can change. The mind is yours. Here is how your team sees the business.

Why visibility matters

An agent making confident decisions on stale data is worse than an agent making no decisions at all. Without live visibility into the state of the business, your team guesses. Guesses turn into wrong calls on pricing, wrong calls on positioning, wrong calls on risk. Bad data makes smart agents look dumb.

Dashboards for every part of your life

Every install gets its own private views. No two SupraOS installations look the same.

↗ Trading Positions, P&L, risk, and active alerts

◎ Research Intelligence feed, sentiment, competitor tracking

♥ My Health Sleep, steps, heart rate, and wellness suggestions

How data stays yours

Every install gets its own private dashboards, scoped to your wallet. A twelve-column drag-and-drop grid with more than thirteen widget types lets you compose the views your business actually needs, with thirteen-plus widget types, five visual skins, and more than forty design tokens. Your agents read the same dashboards you do. They see what you see, and they see nothing else.

Two-tier encryption keeps the data private even from the people running the infrastructure. Your browser encrypts before anything hits the wire. The server decrypts only for the duration of your session, in memory, and never writes the plaintext to disk. The database sees encrypted blobs and nothing more. The root key is derived from your wallet signature, which means you hold it and nobody else can.

Per-agent capability enforcement decides what each agent is allowed to read and do. Sentinel can read every request but can't execute trades. Co-CFO can execute trades but only within the policy you set. Co-CMO can post to your social accounts but can't touch your finance data. The rules are enforced cryptographically, not by promise.

Model freedom keeps your organization intact when the underlying AI changes. Swap Llama for Claude, swap Claude for Gemini, swap cloud for a model running on your own hardware. Prompts get sanitized before they leave your install. State summaries port across providers. The brain can change. The mind, the memory, the team, all stay yours.

Built to be verified

Every agent action is cryptographically signed and hash-chained. Tamper with any record and the chain breaks visibly. Periodic summaries anchor to Supra L1 — proof that lives outside your install and outside ours.

ed25519Cryptographic identity per agent

SHA-256Hash-linked audit chain

AES-256-GCM3-tier encrypted memory

Supra L1On-chain anchoring

YOUR INSTALL · THE INTERNET

●YOUR INSTALL

wallet-derived root key

22 agents, scoped

private dashboards

encrypted memory

model: any, swappable

○THE INTERNET

model providers

data vendors

SaaS dashboards

analytics pipelines

ad platforms

data never crosses the line

Build

How your team actually ships.

A team that can plan but can't execute isn't a team. Yours can do both.

Plans are cheap. Shipped work is the whole job. SupraOS is built around the step most AI products skip: turning a conversation into a finished piece of work. Your team designs visual workflows in a seventy-two-node builder, takes voice commands through Option+Space, and pair-codes with you inside the SupraOS Terminal. Conversations become campaigns. Briefs become code. Decisions become deployed features. You stay in the loop when it matters and out of it when it doesn't. This is where your twenty-two-person organization stops talking about the work and starts doing it.

Why shipping matters

Most AI products stop at the draft. The plan is written, the outline is neat, the strategy is sound, and then a human has to go do all of it. That gap between planning and executing is where weeks disappear. A team that can close that gap is the only kind worth having.

Research that improves itself

Tell the Planner what you need in plain English. It designs the workflow — picking from seventy-two node types across data, logic, action, AI, and integration. Then a Critic Agent scores the output. If it doesn't hit your quality threshold, the system iterates automatically. What lands on your canvas has already been judged good before you see it.

The loop closes without you. Draft → score → iterate → ship. You approve. The Coordinator executes.

Three ways work ships

The visual workflow builder is the first. Describe the outcome, Planner drafts the graph, critics iterate until it passes the bar. You edit, approve, and schedule.

Voice control is the second. Hold Option+Space and talk. The six core platform agents are always listening on your terms, and they know which one of them owns your request before you finish the sentence. Ask Recon for a market sweep, ask Co-CFO for a position update, ask Planner to draft a workflow for a campaign. The command becomes a task, the task becomes work, the work becomes output that lands back in the dashboard you're looking at.

The SupraOS Terminal is the third. It runs paired with your IDE agent, so you can pair-code with a developer that has the whole repo, the whole memory, and the whole team context in view. While the terminal codes the current task, Guide and Planner run in parallel — watching every step, catching scope drift, and pre-writing your next prompt so it's ready the moment the current task lands. Autopilot mode classifies commands by safety, runs the safe ones on its own, and hard-blocks destructive operations entirely. Memory carries across sessions, so tomorrow's Terminal remembers where today's left off. Tasks close with XP and streaks, which feels like a game and keeps the velocity honest. And it wraps any coding agent — Claude, Codex, Cursor, your local model. No lock-in.

HOW WORK SHIPS

STEP 01

>_

Conversation

voice, chat, or paired Terminal

STEP 02

◆◆◆

Workflow

Planner drafts graph, critics iterate

STEP 03

✓

Shipped

deployed, posted, executed

WORKFLOW72 nodes · Planner AI · critic loops until excellent

VOICEOption+Space · 6 core agents respond

TERMINALIDE pair-code · parallel agents · autopilot · cross-session memory · any model

How SupraOS works.