Research and earlier design

Last changed 2026-06-23. This page records research, planning, or an earlier architecture and may not describe the current product. For current behaviour, start at Start here.

AgentChat + Company OS Branding And Marketing

This is the plain-language positioning package for explaining AgentChat and Company OS to nontechnical buyers, partners, operators, investors, and people who assume every AI product is "just ChatGPT with extra buttons."

The Short Version

AgentChat is the AI control room for your company.

You talk to it like a chat app, but behind the chat is Company OS: the memory, task, data, policy, security, reputation, voice, evidence, and learning layer that lets AI agents do real company work without losing control.

The simplest line: ChatGPT helps you think. AgentChat helps your company operate.

The practical line: Turn conversations, voice notes, and company data into verified work handled by AI agents with memory, approvals, evidence, and continuous learning.

The category line: AgentChat is the AI control room powered by Company OS.

Brand Architecture

AgentChat

The human-facing product.

What people see:

• Chat
• Voice
• Tasks
• Agents
• Approvals
• Results
• Evidence
• Follow-up

Plain explanation: AgentChat is where your team talks to the company and turns intent into action.

Use this when talking to nontechnical people. People understand chat. They understand talking. They understand "show me the work." AgentChat is the front door.

Company OS

The serious engine underneath.

What it provides:

• Company memory
• Task ledger and Beads visualization
• Company data access
• SQL/schema understanding
• Agent routing
• Policy and approvals
• Security and reputation
• Evidence trails
• Voice-to-action workflows
• Continuous learning from corrections and outcomes
• File and graph context through ReferenceAtlas-style reference intelligence

Plain explanation: Company OS is the operating layer that gives AI agents memory, permissions, data access, task management, and accountability.

Use this after people understand AgentChat. It gives depth without making the first sentence heavy.

The Core Positioning

Most AI tools are conversations. AgentChat turns conversations into operations.

A normal chatbot answers a prompt. AgentChat can take an instruction, check company data, create tasks, assign work to agents, ask for approval when needed, attach evidence, remember the outcome, and improve the next time.

That is the category difference.

The Best Elevator Pitch

30 seconds

AgentChat is the AI control room for your company. You can talk to it like ChatGPT, but instead of only answering, it connects to how your company actually works. It checks company data, creates tasks, routes work to agents, asks for approval before risky actions, shows evidence, and remembers what happened. So your company does not just use AI. It becomes easier for AI to understand, operate, and improve.

60 seconds

Most companies are trying AI through isolated chats. The problem is that real work needs memory, data, permissions, tasks, approvals, and follow-up. AgentChat gives people the easy front door: they can type or speak what they want. Company OS underneath turns that intent into structured work. It can inspect company data, understand the relevant context, create Beads as a task ledger, route work to the right agent, enforce policy, require human approval when risk is high, record evidence, and learn from corrections. The result is not another chatbot. It is an operating layer for AI work.

Table Pitch

Imagine a manager says: Look at yesterday's customer complaints, find what needs action, draft responses, and ask me before anything goes out.

AgentChat takes that instruction, checks the company's data, finds the risky complaints, creates tasks, assigns agents, drafts the responses, attaches evidence, and pauses before sending. After the manager approves or corrects it, Company OS remembers the pattern. Next time, the company is smarter.

That is the whole product in one story: voice, data, memory, tasks, agents, approvals, evidence, and learning.

What To Say Around A Table

Use this when people are curious but not technical.

Start here: The easiest way to understand it is: AgentChat is not trying to be a smarter chatbot. It is trying to be the place where AI work inside a company gets organized, checked, approved, and remembered.

Then make it concrete: With ChatGPT, you ask a question and get an answer. With AgentChat, you can say, "Check our customer complaints, find what needs action, draft responses, and ask me before anything goes out." Behind the scenes, Company OS turns that into tasks, checks company data, routes agents, attaches evidence, and keeps memory of what happened.

Then name the advantage: The advantage is control. The company gets memory, policy, data access, task management, security, reputation, voice, and evidence around the AI. That is what makes it useful for real operations.

If they ask "what does it actually do?": It turns messy human intent into structured work. A person can speak or type a request; the system can look at company context, create tasks, assign agents, ask approval for risky actions, show evidence, and learn from corrections.

If they ask "why now?": Because companies are moving from "AI that answers" to "AI that acts." Once AI starts acting, you need an operating layer. Otherwise every agent becomes another loose tool with no memory, no accountability, and no shared view of work.

If they ask "what is the buyer buying?": They are buying fewer dropped balls, faster follow-up, safer delegation, and a company that gets easier for AI to operate over time.

The Killer Comparison

Normal AI Chat	AgentChat + Company OS
Answers questions	Runs company work
Knows general knowledge	Knows your company context
One conversation at a time	Turns work into tasks, owners, and follow-up
The user manages everything manually	The system tracks what is waiting, assigned, approved, or blocked
No built-in operational memory	Builds memory from procedures, corrections, outcomes, and evidence
No real accountability	Keeps receipts: tasks, traces, approvals, evidence, and audit history
Risk is handled by prompt discipline	Risk is handled by policy, permissions, approvals, and reputation
Generic intelligence	Company-specific operating intelligence

The short answer to "doesn't ChatGPT already do this?": ChatGPT can help you write or reason. AgentChat can help the company execute because it has the operating layer ChatGPT does not: company memory, tasks, data access, policy, approvals, evidence, and learning.

What To Say To Nontechnical People

Do not lead with modules, papers, acronyms, or internal architecture. Lead with the work they already understand:

• "It helps fewer things fall through the cracks."
• "It remembers how your company works."
• "It checks your company data before acting."
• "It turns requests into visible work."
• "It asks you before risky actions."
• "It shows why it did something."
• "It gets better after corrections."

Then connect the product: You talk to AgentChat. Company OS organizes the work behind the scenes.

What Makes It Different

1. Company Memory

Normal AI starts from the current prompt. Company OS can preserve facts, procedures, corrections, preferences, approvals, task history, and learned patterns.

Nontechnical version: It does not start from zero every time. It learns how your business works.

Why it matters:

• Less repeated explanation
• Better continuity across work
• Faster onboarding for agents and people
• Institutional knowledge becomes usable

2. Company Data

AgentChat is not limited to pasted text. Company OS can connect AI work to company data, schema understanding, retrieval, file context, and graph knowledge.

Nontechnical version: It can look at the business itself, not just guess from a chat.

Why it matters:

• Better answers from real data
• Safer decisions with context
• Less manual exporting and pasting
• More useful analysis for operations, sales, support, finance, and leadership

3. Task Management Through Beads

Work becomes visible. Beads are the task ledger: what exists, what depends on what, who owns it, what evidence is attached, and what is blocked.

Nontechnical version: The chat turns into a workboard, so requests do not disappear.

Why it matters:

• Clear ownership
• Better follow-up
• Less lost work
• Easier review of what agents did
• A bridge between conversation and operations

4. Policy And Approval Gates

Company OS can know when to act, when to pause, what is allowed, and what needs review.

Nontechnical version: It can move fast where safe, and stop for approval where risk is real.

Why it matters:

• Safer delegation
• Better compliance
• Human control over sensitive actions
• Clear separation between suggestions and real-world actions

5. Security And Reputation

The security story is not just "does this role have permission?" It can become dynamic: has this agent, workflow, or tool earned trust for this kind of action?

Nontechnical version: The system can make trust earned, measured, and adjustable.

Why it matters:

• New or unreliable agents can be watched more closely
• Trusted agents can move faster in known lanes
• Failures, corrections, approvals, rejections, and rollbacks become reputation signals
• Reputation can raise or lower approval requirements

6. Evidence And Receipts

AgentChat should not only give a result. It should show the evidence behind important work: source data, task trail, approval, comparison, evaluation, or execution trace.

Nontechnical version: It does not just say "done." It shows the receipt.

Why it matters:

• Easier trust
• Easier audit
• Easier debugging
• Better accountability for agent work

7. Voice And Natural Intent

The user can speak naturally, then the system can clean up the intent, structure it, and turn it into work.

Nontechnical version: You can talk to your company the way you would talk to a chief of staff.

Why it matters:

• Lower friction
• Faster capture of messy ideas
• Better executive/operator workflow
• More natural use in meetings, field work, and daily operations

8. Continuous Learning

Every run can improve future behavior: memory, routing, policy, skills, prompts, and task patterns.

Nontechnical version: The company gets smarter from doing the work.

Why it matters:

• Corrections become reusable
• Procedures become operational memory
• Repeated work becomes easier
• The system compounds instead of resetting

The Simple Mental Models

AI Control Room Best for business owners and operators. Everything your AI agents are doing, remembering, waiting on, and asking approval for lives in one control room.

AI Chief Of Staff Best for founders and executives. It listens, organizes, delegates, follows up, and brings you the decisions that need you.

Company Brain Best for vision. Your company gets a memory, a nervous system, and a way to learn from every task.

Air Traffic Control For AI Agents Best for trust and safety. Agents can move, but the system manages routing, permissions, risk, and approvals.

Work OS For AI Employees Best for category creation. If companies are going to use AI agents, those agents need tasks, memory, permissions, reviews, and managers.

Homepage Direction

Hero

Headline options:

• Run your company through AgentChat.
• The AI control room for company work.
• Turn conversations into company action.
• Talk to your company. Watch work happen.

Recommended headline: Run your company through AgentChat.

Subheadline: Turn conversations, voice notes, and company data into verified work handled by AI agents with memory, approvals, evidence, and continuous learning.

Primary CTA: Open the Control Room

Secondary CTA: See How It Works

Proof strip: Company memory / Task ledger / SQL-aware data / Approval gates / Evidence trail / Learning loop

Avoid

Avoid leading with:

• Module counts
• Paper IDs
• Internal acronyms
• "Multi-agent orchestration" as the first phrase
• "LLM wrapper" language
• Claims that require exact current verification, like live test counts or commit counts

Use technical proof after the buyer understands the outcome.

Demo Story

Use this exact story in meetings:

A business owner opens AgentChat and says, "Look at yesterday's customer complaints, find what needs action, draft responses, and ask me before anything goes out."

AgentChat turns the request into a verified task. Company OS checks the company's data, finds the relevant complaints, creates Beads for the work, assigns the right agent, drafts responses, attaches evidence, and pauses for approval before anything is sent.

If the owner corrects the response, Company OS remembers that correction. The next time a similar complaint appears, the system handles it better.

Then explain: That is the difference. It is not just a conversation. It is memory, data, tasks, agents, approvals, evidence, and learning connected in one loop.

Objection Handling

"Isn't this just ChatGPT?" No. ChatGPT is a powerful general assistant. AgentChat is a company operating surface. ChatGPT can help write an email. AgentChat can check the customer record, create the follow-up task, assign the right agent, draft the email, ask for approval, log the evidence, and remember the correction for next time.

"Why not just use Slack plus ChatGPT?" Slack is where messages happen. ChatGPT is where answers happen. AgentChat is where company work becomes structured, assigned, governed, and remembered. The missing layer is the operating layer: tasks, memory, permissions, approvals, evidence, reputation, and learning.

"How do we trust agents?" You do not trust them blindly. Company OS gives agents lanes. Some work can happen automatically. Sensitive work asks for approval. Risky tools can be gated. Reputation can change how much review is required. Every important result can carry evidence.

"Will this replace people?" The better frame is coordination, leverage, and control. AgentChat helps people delegate work to AI without losing visibility. Humans still set intent, approve sensitive actions, correct behavior, and decide what matters.

"What is the ROI?" The first ROI is fewer dropped balls: requests become tasks, tasks have owners, and follow-up is visible. The second is less repeated context: the system remembers company patterns. The third is safer delegation: agents can do more work while approvals and evidence keep control in place.

"Why would a company need this?" Because once AI moves beyond answering questions, companies need an operating layer. Agents need memory, permissions, tasks, data, approvals, and accountability. Without that, AI stays useful but fragmented.

Sales Conversation Script

Start with the pain: Most AI use inside companies is still trapped in one-off chats. People ask for help, get an answer, then the real work still has to be tracked, checked, approved, and remembered manually.

Introduce the shift: AgentChat changes that. You still get the easy chat interface, but the output becomes company work.

Make it concrete: It can create tasks, check company data, assign agents, ask for approval, show evidence, and remember the result.

Differentiate: The product is not "better prompting." The product is the operating layer behind the prompt.

Close: The goal is simple: let companies use AI agents without losing memory, control, or accountability.

Executive Pitch

Most companies will not win with AI because they bought access to a model. They will win because they made their company understandable and operable by AI. AgentChat is the front door. Company OS is the layer that makes the company legible: memory, tasks, data, policy, permissions, approvals, evidence, reputation, and learning. That is why the product matters. It is not another place to chat. It is the system that lets AI participate in operations.

Investor Pitch

The first wave of AI adoption was personal productivity: answer questions, write content, summarize documents. The next wave is operational AI: agents that do actual company work. But operational AI needs infrastructure that ordinary chat tools do not provide. It needs persistent memory, task state, data access, policy, approvals, evidence, security, reputation, and learning loops. AgentChat is the user-facing control room for that shift. Company OS is the operating substrate underneath. The wedge is familiar: people already know how to chat. The expansion is deep: every conversation can become structured work, memory, data access, approval flows, and agent execution.

Internal Language Rules

Say:

• AI control room
• Company memory
• Task ledger
• Approval gates
• Evidence trail
• Learns from every result
• Company-specific operating intelligence
• Turn conversations into company action

Avoid leading with:

• SEB
• UaC
• HNSW
• SCAR
• HITL
• Multi-agent harness
• Postgres operational substrate
• Eval grader
• Any paper ID

Those are proof points for technical audiences, not first-touch messaging.

Technical Proof, Translated

Product Claim	Technical Backing	Plain-Language Translation
It remembers the company	Unified memory, corrections, task history, learned patterns	It does not start over every conversation
It turns chat into work	Beads ledger, task graph, ownership lifecycle	Requests become visible tasks with owners
It checks company data	SQL/schema tools, connectors, retrieval	It can use the business data, not just guess
It routes work to agents	Agent roles, adaptive context, runtime orchestration	The right kind of agent gets the right kind of work
It asks before risky action	Approval lanes, policy, permissions, certificates	Sensitive work pauses for human review
It shows evidence	Evaluation, comparisons, traces, task history	You can inspect why it did something
It learns from corrections	Correction detection, learning extraction, memory updates	Fixes become future behavior
It can become trust-aware	Reputation signals, tool gating, approval escalation	Trust is earned and can change over time
It understands files and graph context	ReferenceAtlas-style context and graph knowledge integration	It can find relevant project context instead of relying only on chat history
It supports voice-to-action	Voice input, cleanup, intent routing	You can speak messy intent and turn it into work

Best Final Pitch

Most AI tools are conversations. Company OS turns those conversations into operations.

AgentChat is the front door: you talk, type, or dictate what you want. Behind it, Company OS checks company data, creates tasks, assigns agents, applies policy, asks for approval, records evidence, and remembers the result.

That means your company does not just use AI. Your company becomes understandable and operable by AI.