mirror of
https://github.com/tinyhumansai/openhuman.git
synced 2026-07-28 13:32:23 +00:00
GitBook: No commit message
This commit is contained in:
Steven Enamakel
+15
-8
@@ -12,14 +12,16 @@ OpenHuman is the first open-source AI agent with Big Data capabilities and a per
|
||||
|
||||
Every AI model in the world, all 200+ of them, shares the same fundamental limitation. They are stateless. You type a prompt, get a response, and the context evaporates. Even the ones with "memory" store a few bullet points. A few bullet points is a sticky note, not intelligence.
|
||||
|
||||
OpenHuman solves this with two core innovations:
|
||||
OpenHuman solves this with three core innovations:
|
||||
|
||||
<figure><img src=".gitbook/assets/V02 — Two Innovations@2x.png" alt=""><figcaption></figcaption></figure>
|
||||
<figure><img src="/broken/files/YgapS9IPJjw1RqsSmLtW" alt=""><figcaption></figcaption></figure>
|
||||
|
||||
**Neocortex**, a human-like memory engine that can accurately work with over 1 billion tokens. Neocortex indexes 10 million tokens in under 10 seconds, costs $1 to index 5 million tokens, and runs on a MacBook Air CPU with zero LLM dependency. Neocortex goes beyond vector databases. It understands time, entities, and relationships, and it models memory the way the human brain does: by compressing, forgetting strategically, and building knowledge graphs.
|
||||
|
||||
**A personalized subconscious system** inspired by the Purkinje cell, a specialized neuron in the human brain responsible for random thoughts and consciousness. OpenHuman's subconscious triggers over 10,000 memory recall loops per day for under $1, producing proactive insights, pattern recognition, and emergent "thoughts" that feed into a self-learning loop.
|
||||
|
||||
**A local-first desktop app** that runs AI models directly on your machine. OpenHuman uses Gemma 3 for chat, vision, speech-to-text, and text-to-speech, all on-device. Two flagship capabilities power the experience: Screen Intelligence captures and analyzes your screen activity locally, and Auto-complete uses your memory context to suggest completions on any text surface. Your raw data never leaves your device.
|
||||
|
||||
Together, these systems turn OpenHuman into something fundamentally different from a chatbot. It is an AI agent that consumes large amounts of personal data at low cost, maintains a persistent and evolving understanding of your world, and takes proactive actions on your behalf.
|
||||
|
||||
{% hint style="info" %}
|
||||
@@ -30,26 +32,31 @@ OpenHuman is not AGI. But it is a meaningful architectural step closer to it, wi
|
||||
|
||||
OpenHuman connects to your communication platforms, tools, and workflows, and compresses that information into structured intelligence any AI can act on.
|
||||
|
||||
It **watches your screen** through Screen Intelligence, capturing what you see across your apps locally and building a continuous understanding of your work context without sending raw screenshots to any server.
|
||||
|
||||
It **completes your thoughts** through Auto-complete, using your memory context and local models to suggest relevant completions on any text input surface across your system.
|
||||
|
||||
It **compresses context** across your connected sources, turning millions of tokens of organizational noise into a structured knowledge graph of entities, relationships, and temporal chains.
|
||||
|
||||
It **surfaces signals** that matter: decisions, action items, risks, sentiment shifts, and buried context you would otherwise miss.
|
||||
|
||||
It **thinks proactively** through its subconscious system, making connections and surfacing insights you did not explicitly ask for, based on patterns across your entire data history.
|
||||
|
||||
It **preserves privacy** by design. Raw data stays on your device. Encryption keys never leave the device. Only compressed metadata and summaries are processed server-side.
|
||||
It **preserves privacy** by design. Raw data stays on your device. Encryption keys never leave the device. Local models handle sensitive processing on-device. Only compressed metadata and summaries are processed server-side.
|
||||
|
||||
## Who it's for
|
||||
|
||||
OpenHuman is built for people and teams who operate across many conversations and tools, and feel the cost of it.
|
||||
|
||||
- **Developers and power users** in the AI ecosystem, especially those using OpenClaw, CrewAI, and similar agentic frameworks, who need a memory and context layer that actually scales.
|
||||
- **High-volume communicators** who miss decisions, context, and follow-ups buried in message noise across multiple platforms.
|
||||
- **Traders and analysts** who need fast signal extraction and risk awareness across information channels, both on-chain and off-chain.
|
||||
- **Distributed teams** who make decisions in chat but need structured follow-through in external tools.
|
||||
* **Knowledge workers** who spend their days across 8+ applications and lose context every time they switch. Screen Intelligence and Auto-complete turn your daily workflow into cumulative awareness rather than repeated context loss.
|
||||
* **Developers and power users** in the AI ecosystem, especially those using OpenClaw, CrewAI, and similar agentic frameworks, who need a memory and context layer that actually scales.
|
||||
* **High-volume communicators** who miss decisions, context, and follow-ups buried in message noise across multiple platforms.
|
||||
* **Traders and analysts** who need fast signal extraction and risk awareness across information channels, both on-chain and off-chain.
|
||||
* **Distributed teams** who make decisions in chat but need structured follow-through in external tools.
|
||||
|
||||
## What OpenHuman does not do
|
||||
|
||||
OpenHuman does not claim to be AGI. It does not send messages automatically or impersonate you. It does not store your raw message data long-term. It does not train on your data.
|
||||
OpenHuman does not claim to be AGI. It does not take actions in your connected platforms without your explicit instruction. It does not store your raw message data long-term. It does not train on your data.
|
||||
|
||||
{% hint style="info" %}
|
||||
Privacy is a core architectural decision, not a checkbox. The full privacy design is covered in Privacy & Security.
|
||||
|
||||
@@ -8,24 +8,24 @@ This section walks you through setting up OpenHuman and running your first reque
|
||||
|
||||
OpenHuman is open source under the GNU GPL3 license. The codebase is available at [github.com/tinyhumansai/openhuman](https://github.com/tinyhumansai/openhuman). You can self-host, contribute, or simply use the hosted version.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Download & Install
|
||||
|
||||
OpenHuman runs natively on:
|
||||
Download the OpenHuman desktop app from the official website at [tryopenhuman.com](https://tryopenhuman.com). The desktop app runs local AI models on your machine for private screen analysis, auto-complete, and chat.
|
||||
|
||||
- **macOS** (Intel and Apple Silicon)
|
||||
- **Windows** (x64 and ARM64)
|
||||
- **Linux** (x64 and ARM64: AppImage and .deb)
|
||||
- **Android**
|
||||
- **iOS**
|
||||
- **Web** (any modern browser)
|
||||
OpenHuman runs natively on
|
||||
|
||||
Download the app for your platform from the official OpenHuman website, or use the web version directly in your browser.
|
||||
* **macOS** (Intel and Apple Silicon)
|
||||
* **Windows** (x64 and ARM64)
|
||||
* **Linux** (x64 and ARM64: AppImage and .deb)
|
||||
* **Android**
|
||||
*  **iOS**
|
||||
* **Web** (any modern browser).
|
||||
|
||||
The native app has a small footprint, starts fast, and uses your operating system's secure credential storage. The web version provides the same core functionality when a native install is not practical.
|
||||
The desktop app is the primary experience. It has a small footprint, starts fast, runs local AI models, and uses your operating system's secure credential storage. Mobile and web versions provide companion access to your account and intelligence layer when you are away from your desktop.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Create Your Account
|
||||
|
||||
@@ -37,7 +37,29 @@ After signing in, you may be placed on a waitlist depending on rollout status. T
|
||||
**No permanent lock-in.** Creating an account does not grant OpenHuman ongoing access to anything. All processing still requires explicit actions from you later.
|
||||
{% endhint %}
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
### Set Up Screen Intelligence
|
||||
|
||||
Screen Intelligence is one of OpenHuman's core capabilities. It captures your screen activity locally and builds context from what you see across your applications throughout the day.
|
||||
|
||||
**How it works:** The desktop app takes screenshots approximately every 5 seconds and processes them using the on-device vision model (Gemma 3). Each capture is analyzed into a structured summary: which app you were using, what content was on screen, and what you were doing. The raw screenshots are processed and discarded. Only the compressed summaries persist in your local Neocortex memory.
|
||||
|
||||
**Per-app permissions:** You control which applications Screen Intelligence monitors. When you first enable it, you can review and adjust which apps are included. Exclude any application you want to keep private, such as banking, medical, or personal apps. You can change these permissions at any time from settings.
|
||||
|
||||
**What gets stored:** Structured summaries only. "User was reviewing a spreadsheet in Google Sheets showing Q3 revenue figures" rather than a pixel-for-pixel copy of your screen. Raw screenshots never leave your device and are not stored after processing.
|
||||
|
||||
***
|
||||
|
||||
### Enable Auto-complete
|
||||
|
||||
Auto-complete uses your Neocortex memory combined with the local model to suggest relevant text completions on any input surface across your system.
|
||||
|
||||
Because it draws on your accumulated context, suggestions reflect your actual terminology, projects, contacts, and patterns. This is different from generic auto-complete: it knows what you have been working on, who you have been talking to, and what is currently relevant.
|
||||
|
||||
Auto-complete runs entirely on your local model. No keystroke data or text input is sent to any server. You can enable or disable it from settings at any time.
|
||||
|
||||
***
|
||||
|
||||
## Connect Your First Source
|
||||
|
||||
@@ -51,36 +73,51 @@ OpenHuman works by connecting to your existing tools and platforms. Each connect
|
||||
|
||||
**Slack:** Connect Slack to extend your knowledge graph across workplace conversations.
|
||||
|
||||
All integrations are optional. You can start with one source and add more later. The more sources you connect, the more powerful the intelligence becomes. Each connection is independently revocable.
|
||||
All integrations are optional. You can start with Screen Intelligence alone, without connecting any external source, and add integrations later. The more sources you connect, the more powerful the intelligence becomes. Each connection is independently revocable.
|
||||
|
||||
---
|
||||
> **Planned integrations:** Discord, iMessage, WhatsApp, Gmail, Google Calendar, and Web3 wallets are on the roadmap.
|
||||
|
||||
***
|
||||
|
||||
## Run Your First Request
|
||||
|
||||
Once a source is connected, you can start asking OpenHuman questions using the command input.
|
||||
Once Screen Intelligence is running or a source is connected, you can start asking OpenHuman questions.
|
||||
|
||||
Try prompts like:
|
||||
|
||||
- **"Summarize what I missed today"**
|
||||
- **"What are the key decisions from this week?"**
|
||||
- **"Extract action items from my recent conversations"**
|
||||
- **"What topics are trending across my groups?"**
|
||||
**Screen-aware queries:**
|
||||
|
||||
* "What was I working on this morning?"
|
||||
* "Summarize what I saw in that meeting"
|
||||
* "What was on that dashboard I was looking at earlier?"
|
||||
|
||||
**Messaging queries:**
|
||||
|
||||
* "Summarize what I missed today"
|
||||
* "What are the key decisions from this week?"
|
||||
* "Extract action items from my recent conversations"
|
||||
* "What topics are trending across my groups?"
|
||||
|
||||
**Cross-source queries:**
|
||||
|
||||
* "Connect what my team discussed in Slack with what I was reviewing on screen"
|
||||
* "What did Sarah say about the project I was working on yesterday?"
|
||||
|
||||
OpenHuman processes only the data needed to answer your request, produces an output, and presents it for your review. If the output feels too broad or too shallow, narrow the scope. Specify a particular conversation, time range, or intent.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Explore Skills & Integrations
|
||||
|
||||
After your first request, explore what else OpenHuman can do:
|
||||
|
||||
- **Skills** extend the assistant's capabilities, fetching external data, running scheduled tasks, processing information, and writing outputs to connected tools
|
||||
- **Integrations** let you push structured results to Notion, Google Sheets, and other tools
|
||||
- **Workflows** turn conversation decisions into trackable actions you can follow through on
|
||||
* **Skills** extend the assistant's capabilities, fetching external data, running scheduled tasks, processing information, and writing outputs to connected tools
|
||||
* **Integrations** let you push structured results to Notion, Google Sheets, and other tools
|
||||
* **Workflows** turn conversation decisions into trackable actions you can follow through on
|
||||
|
||||
Learn more in [Skills & Integrations](../product/skills-and-integrations.md).
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### Join the Community
|
||||
|
||||
|
||||
@@ -14,9 +14,21 @@ Research from Google's Titans project demonstrates the core issue clearly. As LL
|
||||
|
||||
<figure><img src="../.gitbook/assets/V04_The_ANI_Problem@2x.png" alt=""><figcaption></figcaption></figure>
|
||||
|
||||
OpenHuman solves this in two steps.
|
||||
OpenHuman solves this in three steps.
|
||||
|
||||
#### Step 1: Neocortex, the memory engine
|
||||
#### **Step 1: A local-first desktop app**
|
||||
|
||||
Before anything reaches the cloud, OpenHuman runs AI models directly on your machine. The desktop app uses Gemma 3 for chat, vision analysis, speech-to-text, and text-to-speech, all running locally on your device's hardware.
|
||||
|
||||
Two capabilities define the desktop experience:
|
||||
|
||||
**Screen Intelligence** captures screenshots of your screen approximately every 5 seconds and processes them locally using the on-device vision model. Each capture is summarized into structured context: what application you were using, what content was visible, what actions you were taking. Raw screenshots are not stored long-term or sent to any server. You control which apps are included through per-app permissions, so you can exclude sensitive applications like banking or medical tools.
|
||||
|
||||
**Auto-complete** uses your accumulated memory context combined with the local model to generate relevant text completions on any input surface across your system. Because it draws on your Neocortex memory, the suggestions reflect your actual work context, terminology, and patterns rather than generic predictions.
|
||||
|
||||
Both features run entirely on-device. No raw screen data or keystroke data touches any server.
|
||||
|
||||
#### Step 2: Neocortex, the memory engine
|
||||
|
||||
Neocortex is a human-like AI memory system that can accurately work with over 1 billion tokens. It is the foundation everything else is built on.
|
||||
|
||||
@@ -48,7 +60,7 @@ Additional capabilities of Neocortex:
|
||||
|
||||
**Temporal context weighting** uses time-decay functions so recent events carry more weight than older ones, matching how human attention naturally works.
|
||||
|
||||
#### Step 2: The personalized subconscious
|
||||
#### Step 3: The personalized subconscious
|
||||
|
||||
With good memory and good recall, OpenHuman can do something no other AI agent does: maintain a personalized subconscious.
|
||||
|
||||
@@ -66,17 +78,21 @@ The success metric for the subconscious is what the team calls the "mirror test"
|
||||
|
||||
When you use OpenHuman, here is what happens:
|
||||
|
||||
<figure><img src="../.gitbook/assets/V08_How_Pieces_Connect@2x.png" alt=""><figcaption></figcaption></figure>
|
||||
<figure><img src="/broken/files/qZp0NB9a68hME51I4bLX" alt=""><figcaption></figcaption></figure>
|
||||
|
||||
**Your desktop app runs locally.** Screen Intelligence captures your screen context. Auto-complete assists your typing. Local models handle chat, vision, and voice. All of this runs on your device with no server dependency.
|
||||
|
||||
<figure><img src="/broken/files/cVwq7j0D8bm7POwV4L1V" alt=""><figcaption></figcaption></figure>
|
||||
|
||||
**Connect your data sources.** Telegram, Slack, Gmail, Notion, blockchain wallets, and more. Each connection expands your knowledge graph.
|
||||
|
||||
**Neocortex compresses.** Millions of tokens of organizational history become structured intelligence: entities, relationships, timelines, sentiment. Noise is stripped. Signal is preserved.
|
||||
**Neocortex compresses.** Millions of tokens of organizational history, including both your screen activity summaries and your messaging data, become structured intelligence: entities, relationships, timelines, sentiment. Noise is stripped. Signal is preserved.
|
||||
|
||||
**The subconscious runs.** Thousands of recall loops per day surface proactive insights, track evolving patterns, and update the knowledge graph.
|
||||
|
||||
**You interact naturally.** Ask anything about your life, your team, your projects. Not general knowledge. Your knowledge. What did your team decide? What is the status? What did you miss? OpenHuman already knows.
|
||||
**You interact naturally.** Ask anything about your life, your team, your projects. Not general knowledge. Your knowledge. What did your team decide? What is the status? What did you miss? What was on that dashboard you were looking at this morning? OpenHuman already knows.
|
||||
|
||||
**Privacy is maintained.** Raw data stays on device, encrypted with AES-256-GCM. Encryption keys never leave the device. Only compressed metadata and summaries are processed server-side.
|
||||
**Privacy is maintained.** Raw data stays on device, encrypted with AES-256-GCM. Encryption keys never leave the device. Screen captures are processed locally and discarded. Only compressed metadata and summaries are processed server-side.
|
||||
|
||||
### The architecture: OpenClaw foundation
|
||||
|
||||
@@ -90,6 +106,8 @@ All agents share a common compressed context provided by Neocortex. This effecti
|
||||
|
||||
OpenHuman works with probabilistic models. It may occasionally miss nuance, misinterpret sarcasm, or over-prioritize certain messages. This is more likely in highly informal conversations, fast-moving threads, or contexts with limited data.
|
||||
|
||||
Screen Intelligence processes visual information through a local vision model, which means accuracy depends on screen clarity, text size, and application complexity. Highly dynamic or visually dense interfaces may produce less precise summaries.
|
||||
|
||||
OpenHuman is not AGI. It is a meaningful step closer, with better memory and better orchestration for memory, taking inspiration from the human brain. But human judgment still matters. OpenHuman helps you think faster. It does not think for you.
|
||||
|
||||
Everything is in early alpha. Feedback and contributions are welcomed.
|
||||
|
||||
@@ -6,7 +6,7 @@ icon: layer-plus
|
||||
|
||||
OpenHuman is a native application that runs on six platforms from a single codebase. It is not a web-only tool, browser extension, or Electron wrapper. It is built for performance, security, and a small footprint on every device you use.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Six Platforms, One Experience
|
||||
|
||||
@@ -23,7 +23,21 @@ OpenHuman compiles to native binaries for each supported platform:
|
||||
|
||||
Your account, connected sources, preferences, and settings sync across all platforms. You can start a request on your desktop and review the output on your phone.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Desktop-First Capabilities
|
||||
|
||||
The desktop app is the primary OpenHuman experience. It runs local AI models on your machine and provides capabilities that are only available on desktop:
|
||||
|
||||
**Screen Intelligence** captures your screen approximately every 5 seconds, processes captures through the on-device vision model, and produces structured context summaries. This feature requires the desktop app because it needs access to your screen and local compute for real-time processing. Per-app permissions let you control which applications are monitored.
|
||||
|
||||
**Auto-complete** uses your Neocortex memory context combined with the local model to generate relevant text completions on any input surface. This runs entirely on-device and is available on desktop platforms.
|
||||
|
||||
**Local model inference** for chat, vision, speech-to-text, and text-to-speech runs on your desktop hardware. No GPU is required, though GPU acceleration is used when available.
|
||||
|
||||
Mobile and web versions provide full access to your Neocortex memory, connected source intelligence, and subconscious insights, but do not run local models or Screen Intelligence.
|
||||
|
||||
***
|
||||
|
||||
## Why Native Matters
|
||||
|
||||
@@ -35,13 +49,13 @@ OpenHuman is built as a native application rather than a web wrapper for three r
|
||||
|
||||
**OS-level security.** On desktop platforms, OpenHuman stores credentials in your operating system's secure keychain (macOS Keychain, Windows Credential Manager, Linux Secret Service). Sensitive data never sits in browser storage or plain text files.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Architecture at a Glance
|
||||
|
||||
OpenHuman operates across three layers:
|
||||
|
||||
**Application layer.** The native app on your device handles the interface, user input, local state, credential management, and skill execution. This layer is responsible for everything you see and interact with.
|
||||
**Application layer.** The native app on your device handles the interface, user input, local state, credential management, local model inference, Screen Intelligence, Auto-complete, and skill execution. This layer is responsible for everything you see and interact with, and for all on-device AI processing.
|
||||
|
||||
**Intelligence layer.** OpenHuman's analysis, coordination, and intelligence systems run as a secure backend service. When a request requires deeper language processing, it is handled here. This layer is operated and maintained by OpenHuman.
|
||||
|
||||
@@ -51,7 +65,7 @@ OpenHuman operates across three layers:
|
||||
The intelligence layer is not part of the client application. It performs analysis, coordination, and trust scoring separately from the frontend.
|
||||
{% endhint %}
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Real-Time Communication
|
||||
|
||||
@@ -59,10 +73,12 @@ OpenHuman maintains a persistent connection between the application and the inte
|
||||
|
||||
The connection is designed for resilience. If the network drops, OpenHuman reconnects automatically with progressive backoff. There is no manual reconnection process.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Offline Behavior
|
||||
|
||||
OpenHuman's local state persists on your device. Your preferences, settings, and connected source configurations remain available even when you are offline.
|
||||
|
||||
Full analysis and intelligence features require a network connection, since they depend on the intelligence layer. When connectivity is restored, the app resumes normal operation without requiring you to re-authenticate or reconfigure.
|
||||
On desktop, local model capabilities continue to function offline. Screen Intelligence continues capturing and processing locally. Auto-complete continues generating suggestions from cached memory context. Chat with the local model works without a network connection.
|
||||
|
||||
Full analysis and intelligence features that require the server-side intelligence layer need a network connection. When connectivity is restored, the app resumes normal operation without requiring you to re-authenticate or reconfigure.
|
||||
|
||||
+21
-33
@@ -4,56 +4,44 @@ icon: sack-dollar
|
||||
|
||||
# Pricing
|
||||
|
||||
OpenHuman is priced as a SaaS product for individuals and teams that rely on high-volume communication and need clarity, coordination, and safety across their tools. Pricing reflects the value of reduced cognitive load and better signal quality, not raw message volume.
|
||||
OpenHuman is currently in early alpha. During this phase, access is free for early users.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Individual Plans
|
||||
### Alpha Access
|
||||
|
||||
Individual plans are designed for power users who want help managing high-volume activity across multiple conversations and connected sources.
|
||||
All features are available at no cost during the alpha period, including Screen Intelligence, Auto-complete, Telegram integration, Notion and Google Sheets exports, and subconscious intelligence.
|
||||
|
||||
Core analysis functionality is included so users can understand OpenHuman's value before committing. Deeper features, longer context windows, persistent workflows, richer integrations, and advanced intelligence are available in higher tiers.
|
||||
Early users who join during alpha will receive advance notice before paid plans take effect, along with a transition period and loyalty benefits.
|
||||
|
||||
Plans are tiered to allow users to start lightweight and expand as their needs grow.
|
||||
***
|
||||
|
||||
---
|
||||
### Planned Pricing Model
|
||||
|
||||
## Team Plans
|
||||
OpenHuman will be priced as a subscription for individuals and teams.
|
||||
|
||||
Team plans are designed for shared coordination environments where multiple members need access under a single workspace.
|
||||
**Free tier:** Core functionality with local models. Screen Intelligence, Auto-complete, and basic Neocortex memory using on-device processing.
|
||||
|
||||
Team plans include shared workflows, community-level intelligence, coordination support, and centralized billing. Pricing reflects the number of active participants and the coordination value created.
|
||||
**Paid subscription:** Access to premium cloud models for deeper analysis, full Neocortex intelligence at scale, unlimited subconscious processing, priority support, and advanced integrations.
|
||||
|
||||
Individual user privacy is preserved even in shared team environments.
|
||||
Pricing details and tier breakdowns will be published before the alpha period ends.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Credit System
|
||||
### Team Plans
|
||||
|
||||
OpenHuman uses a credit-based system for usage-metered features.
|
||||
Team plans will support shared coordination environments where multiple members need access under a single workspace. Team plans will include shared workflows, community-level intelligence, coordination support, and centralized billing.
|
||||
|
||||
- Credits are consumed when you use analysis, intelligence, and processing features
|
||||
- Oldest credits are consumed first, encouraging usage before expiry
|
||||
- Credits can be earned through the referral program, in addition to purchase
|
||||
- Every credit transaction is tracked in an auditable ledger
|
||||
***
|
||||
|
||||
Credits provide flexibility as you pay for what you use, and you can supplement your plan with earned credits from referrals.
|
||||
### Credit System
|
||||
|
||||
---
|
||||
OpenHuman uses a credit-based system for usage-metered features beyond the subscription base. Credits are consumed when you use analysis, intelligence, and processing features. Credits can be earned through the referral program, in addition to purchase.
|
||||
|
||||
## Payment Methods
|
||||
***
|
||||
|
||||
OpenHuman supports standard and cryptocurrency payment options.
|
||||
### Payment Methods
|
||||
|
||||
- **Card payments:** Standard subscription billing with the ability to upgrade, downgrade, or cancel at any time
|
||||
- **Cryptocurrency:** Alternative payment path for users who prefer crypto
|
||||
**Card payments:** Standard subscription billing with the ability to upgrade, downgrade, or cancel at any time.
|
||||
|
||||
Changes to your plan take effect immediately. Cancellation does not affect outputs already exported to external tools.
|
||||
|
||||
---
|
||||
|
||||
## Access During Launch
|
||||
|
||||
During the initial launch phase, access to OpenHuman may be gated through a waitlist. This is done to manage platform stability and collect feedback from early users.
|
||||
|
||||
Waitlist access does not imply long-term pricing commitments. Pricing and plan boundaries may evolve during this phase as usage patterns become clearer.
|
||||
**Cryptocurrency:** Alternative payment path for users who prefer crypto.
|
||||
|
||||
@@ -6,21 +6,41 @@ icon: shield
|
||||
|
||||
OpenHuman operates on a principle of **zero-knowledge intelligence**. The system is architecturally designed so that your raw data never needs to leave your device. Neocortex compresses your data locally into structured metadata and summaries. Only this compressed output is processed server-side. Your AI has months of context about your entire organizational life. Your raw data has never touched our servers.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Privacy by Design
|
||||
|
||||
OpenHuman operates on a principle of **zero retention** for message content. When you make a request, the relevant data is processed to produce an output, and the source content is discarded afterward.
|
||||
|
||||
**No long-term message storage.** OpenHuman does not maintain a persistent archive of your conversations. Context is reconstructed from your connected sources when needed.
|
||||
**No long-term raw data storage.** OpenHuman does not maintain a persistent archive of your conversations or screen captures. Context is compressed into knowledge graph structures. Raw text and raw images are not stored.
|
||||
|
||||
**No training on your data.** Your conversations, analysis results, and personal information are never used to train AI models or improve systems. Your data serves you and only you.
|
||||
|
||||
**OS-level credential storage.** On desktop platforms, OpenHuman uses your operating system's secure keychain to store credentials and sensitive tokens. Credentials are never stored in plain text, browser storage, or application-level databases.
|
||||
|
||||
**On-device where possible.** Interface rendering, input handling, local state management, and credential operations all happen on your device. Only tasks requiring deeper language processing are handled server-side, under the same privacy constraints.
|
||||
**On-device where possible.** Screen Intelligence, Auto-complete, local chat, voice processing, and encryption all happen on your device. Only tasks requiring deeper language processing beyond local model capabilities are handled server-side, under the same privacy constraints.
|
||||
|
||||
---
|
||||
### Screen Intelligence Privacy
|
||||
|
||||
Screen Intelligence captures your screen approximately every 5 seconds and processes captures using the on-device vision model. This is the most privacy-sensitive feature in OpenHuman, and it is designed accordingly.
|
||||
|
||||
**Processing is entirely local.** Raw screenshots are analyzed by the local vision model running on your device's hardware. The model produces structured text summaries. The raw screenshot is then discarded.
|
||||
|
||||
**Raw screenshots are never stored.** After the on-device model processes a capture, the raw image is deleted. Only the compressed text summary persists in your local Neocortex memory.
|
||||
|
||||
**Raw screenshots are never transmitted.** No raw screen image data is ever sent to any server, cloud service, or third party.
|
||||
|
||||
**Per-app permissions.** You control which applications Screen Intelligence monitors. You can exclude any application from capture: banking apps, medical portals, password managers, or any application you consider sensitive.
|
||||
|
||||
**You can disable it entirely.** Screen Intelligence is optional. You can turn it off at any time from settings.
|
||||
|
||||
***
|
||||
|
||||
### Auto-complete Privacy
|
||||
|
||||
Auto-complete runs on the local model using cached memory context from Neocortex. No keystroke data, text input, or completion suggestions are sent to any server. The entire feature operates on-device.
|
||||
|
||||
***
|
||||
|
||||
## Permissions and Access Control
|
||||
|
||||
@@ -36,9 +56,10 @@ OpenHuman does not silently expand its access over time. There is no progressive
|
||||
|
||||
Each connected source has its own permission scope:
|
||||
|
||||
- **Telegram:** Read-only access. OpenHuman can read messages from conversations you reference in a request. It cannot send messages, edit messages, react, join groups, or act on your behalf.
|
||||
- **Notion:** Write access to specific workspaces or pages you approve. OpenHuman does not read unrelated documents.
|
||||
- **Google Sheets:** Write access to specific spreadsheets you approve. OpenHuman does not read unrelated sheets.
|
||||
* **Telegram:** Full capability access, including reading messages, sending messages, managing contacts, and administering groups. All actions are executed only on your explicit instruction. OpenHuman never sends messages, joins groups, or takes any Telegram action autonomously. You remain in control of every action. See Skills & Integrations for the complete capability list.
|
||||
* **Notion:** Write access to specific workspaces or pages you approve. OpenHuman does not read unrelated documents.
|
||||
* **Google Sheets:** Write access to specific spreadsheets you approve. OpenHuman does not read unrelated sheets.
|
||||
* **Slack:** Read access to conversations you reference in your requests. Used to expand your knowledge graph with workplace context.
|
||||
|
||||
Integration permissions are limited to what is needed for the specific action you request.
|
||||
|
||||
@@ -46,25 +67,27 @@ Integration permissions are limited to what is needed for the specific action yo
|
||||
|
||||
Every meaningful operation in OpenHuman is user-initiated. Summaries, analysis, trust evaluation, workflow creation, and exports all require a direct request. There is no continuous background processing or monitoring.
|
||||
|
||||
The one exception is Screen Intelligence, which runs on a continuous capture cycle when enabled. However, Screen Intelligence operates entirely on-device and does not access any connected platform. It captures what is visible on your screen.
|
||||
|
||||
{% hint style="info" %}
|
||||
OpenHuman is idle unless you ask it to do something.
|
||||
OpenHuman is idle with respect to your connected platforms unless you ask it to do something.
|
||||
{% endhint %}
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Revoking Access
|
||||
|
||||
You can revoke OpenHuman's access to any connected source at any time.
|
||||
|
||||
- Disconnect a source from your settings
|
||||
- Remove integration permissions
|
||||
- Stop using the application entirely
|
||||
* Disconnect a source from your settings
|
||||
* Remove integration permissions
|
||||
* Stop using the application entirely
|
||||
|
||||
Once access is revoked, OpenHuman immediately stops processing data from that source. There is no delayed or cached processing after revocation. Previously exported outputs (such as summaries written to Notion or Google Sheets) remain where they were written, but no new processing occurs.
|
||||
|
||||
This makes OpenHuman safe to test, pause, or stop using without residual exposure.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Security
|
||||
|
||||
@@ -80,7 +103,7 @@ OpenHuman implements security at every layer of the system.
|
||||
|
||||
**Short-lived tokens.** Authentication tokens are time-limited and single-use where applicable, reducing the window of exposure if a token is compromised.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## How Neocortex enables privacy
|
||||
|
||||
@@ -108,13 +131,13 @@ OpenHuman includes an intelligence layer designed to help you reason about credi
|
||||
|
||||
Trust and risk intelligence operates at different levels:
|
||||
|
||||
- **Personal:** Visible only to you. Your own analysis, trust assessments, and risk alerts.
|
||||
- **Community:** Aggregated patterns within a group or organization, supporting shared coordination and moderation. Never exposes individual message content.
|
||||
- **Network:** Anonymized patterns across the broader OpenHuman user base, improving early detection of shared risks like recurring scam vectors. 
|
||||
* **Personal:** Visible only to you. Your own analysis, trust assessments, and risk alerts.
|
||||
* **Community:** Aggregated patterns within a group or organization, supporting shared coordination and moderation. Never exposes individual message content.
|
||||
* **Network:** Anonymized patterns across the broader OpenHuman user base, improving early detection of shared risks like recurring scam vectors.
|
||||
|
||||
Information does not move between scopes without abstraction and anonymization.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Shared Environments
|
||||
|
||||
|
||||
+63
-31
@@ -8,49 +8,67 @@ icon: question
|
||||
|
||||
#### What is OpenHuman?
|
||||
|
||||
OpenHuman is a personal AI assistant that connects to your communication platforms and productivity tools. It helps you manage high-volume conversations by summarizing, extracting signals, suggesting responses, and creating structured workflows all from a native app that runs on your device.
|
||||
OpenHuman is a personal AI assistant that runs local AI models on your device and connects to your communication platforms and productivity tools. It helps you build continuous awareness across your digital life through Screen Intelligence, Auto-complete, conversation analysis, signal extraction, and structured workflows.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### What is Neocortex
|
||||
|
||||
Neocortex is OpenHuman's memory engine. It is a human-like AI memory system that can work with over 1 billion tokens of data. It indexes 10 million tokens in under 10 seconds, costs $1 to index 5 million tokens, and runs on a MacBook Air CPU with no GPU required. Unlike vector databases, Neocortex understands time, entities, and relationships. It builds knowledge graphs and manages memory through tiered compression inspired by how the human brain works. Learn more in Neocortex.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### What does "Big Data AI" mean?
|
||||
|
||||
Every AI model today is a prompt engine. You type something, it responds, and the context disappears. OpenHuman is different. It compresses your entire organizational life, messages, documents, tools, transactions, into a structured knowledge graph that persists and evolves. This is what we mean by Big Data AI: an AI that operates on months of your real data, not just the prompt you typed right now.
|
||||
|
||||
**How is OpenHuman different from ChatGPT, Claude, or Gemini?**
|
||||
#### **How is OpenHuman different from ChatGPT, Claude, or Gemini?**
|
||||
|
||||
Those models are brilliant at reasoning and generation. But they are stateless. They know nothing about your actual life beyond what you paste into the chat window. OpenHuman is the context layer that makes those models useful. It compresses your organizational data into structured intelligence that any AI can reason over. Think of it this way: ChatGPT is the brain. OpenHuman is the memory.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
**How is OpenHuman different from other AI memory solutions like Mem0, SuperMemory, or MemGPT?**
|
||||
#### **How is OpenHuman different from other AI memory solutions like Mem0, SuperMemory, or MemGPT?**
|
||||
|
||||
Most AI memory solutions use vector databases that retrieve whatever is semantically similar, but similarity alone says nothing about importance. They also cannot support consciousness-like systems or process data accurately at scale beyond 10 million tokens. Neocortex is architecturally different: it uses tiered memory, knowledge graphs, temporal weighting, and semantic deduplication. It processes 10 million tokens in under 10 seconds and supports over 1 billion tokens total. It does this with zero LLM dependency.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
**Is OpenHuman open source?**
|
||||
#### **Is OpenHuman open source?**
|
||||
|
||||
Yes. OpenHuman is built on the OpenClaw architecture and licensed under GNU GPL3. The full codebase is available on [GitHub](https://github.com/tinyhumansai/openhuman). Neocortex benchmarks are also open-sourced. Contributions and feedback are welcomed.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
**Is OpenHuman AGI?**
|
||||
#### **Is OpenHuman AGI?**
|
||||
|
||||
No. OpenHuman is not AGI, and we do not claim it is. It is a meaningful architectural step closer to AGI, with innovations in memory (Neocortex) and consciousness-like processing (the subconscious system) that go beyond what existing agentic systems offer. But it operates within defined boundaries and requires human judgment.
|
||||
|
||||
***
|
||||
|
||||
#### **Does Screen Intelligence see everything on my screen?**
|
||||
|
||||
Only for the apps you allow. Screen Intelligence captures screenshots approximately every 5 seconds and processes them using the on-device vision model. You control which applications are included through per-app permissions. You can exclude any application, such as banking, medical, or personal apps. Raw screenshots are processed locally and discarded. Only compressed text summaries are stored. No raw screen images are ever sent to any server.
|
||||
|
||||
***
|
||||
|
||||
#### **Can I turn off Screen Intelligence?**
|
||||
|
||||
Yes. Screen Intelligence is entirely optional. You can disable it at any time from settings. OpenHuman continues to function with your other connected sources and manual queries.
|
||||
|
||||
***
|
||||
|
||||
#### **Does Auto-complete send my typing to a server?**
|
||||
|
||||
No. Auto-complete runs entirely on the local model using cached memory context from Neocortex. No keystroke data, text input, or completion suggestions leave your device.
|
||||
|
||||
#### Does OpenHuman read all my messages?
|
||||
|
||||
No. OpenHuman only processes messages **when you ask it to** and only within the scope of your request.
|
||||
|
||||
If you ask it to summarize a specific conversation, it reads that conversation. If you do not reference a source, it is not accessed. There is no background monitoring or continuous scanning.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### Is my data safe?
|
||||
|
||||
@@ -58,44 +76,44 @@ Yes. OpenHuman is designed around zero retention of message content. Data is pro
|
||||
|
||||
On desktop platforms, credentials are stored in your operating system's secure keychain. All communication between the app and OpenHuman's servers is encrypted.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### Does OpenHuman store my messages?
|
||||
|
||||
Messages are processed only to fulfill your request. They are **not permanently stored or reused** beyond producing the requested output. Derived intelligence like summaries and workflow records may persist, but raw message content does not.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### Can OpenHuman send messages on my behalf?
|
||||
|
||||
No. OpenHuman does not auto-send messages, post in your groups, or act on your behalf inside any connected platform. If a reply is suggested, you choose whether to use it.
|
||||
OpenHuman can send messages, reply, forward, and take other actions in your Telegram account, but only when you explicitly instruct it to. It never sends messages autonomously or acts without your direct request.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### Who is OpenHuman for?
|
||||
|
||||
OpenHuman is useful for anyone who:
|
||||
|
||||
- Manages high-volume communication across multiple platforms and groups
|
||||
- Needs to stay on top of decisions, action items, and context without reading everything
|
||||
- Works in distributed teams, communities, or coordination-heavy environments
|
||||
- Wants structured outputs from conversations, exportable to tools like Notion or Google Sheets
|
||||
* Manages high-volume communication across multiple platforms and groups
|
||||
* Needs to stay on top of decisions, action items, and context without reading everything
|
||||
* Works in distributed teams, communities, or coordination-heavy environments
|
||||
* Wants structured outputs from conversations, exportable to tools like Notion or Google Sheets
|
||||
|
||||
You do not need to be technical to use it.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### What platforms does OpenHuman support?
|
||||
|
||||
OpenHuman runs natively on **macOS, Windows, Linux, Android, and iOS**, with a **web version** for browser access. Your account and settings sync across all platforms.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### What integrations are available?
|
||||
|
||||
OpenHuman currently integrates with **Telegram** (read and analyze conversations), **Notion** (export structured outputs), and **Google Sheets** (export tabular data and reports). More integrations are planned.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### How much does OpenHuman cost?
|
||||
|
||||
@@ -103,7 +121,7 @@ OpenHuman offers individual and team plans with core analysis included. Deeper f
|
||||
|
||||
See [Pricing](../product/pricing.md) for details.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Troubleshooting
|
||||
|
||||
@@ -113,7 +131,7 @@ If a summary feels incomplete, the most common cause is overly broad scope. When
|
||||
|
||||
**Solution:** Narrow the request to a specific conversation, time window, or intent. OpenHuman performs best when it knows what kind of outcome you are looking for.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### Important context seems missing
|
||||
|
||||
@@ -121,7 +139,7 @@ OpenHuman only processes the data required to fulfill a request. If relevant con
|
||||
|
||||
**Solution:** Expand the scope explicitly by referencing additional conversations or extending the time range.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### Outputs feel incorrect or misinterpreted
|
||||
|
||||
@@ -129,7 +147,21 @@ OpenHuman interprets conversations probabilistically. Tone, sarcasm, and informa
|
||||
|
||||
**Solution:** Refine the request or re-run analysis with a narrower scope. Outputs should be treated as assistance rather than ground truth.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### **Screen Intelligence summaries seem inaccurate**
|
||||
|
||||
The on-device vision model processes screenshots locally. Accuracy depends on screen clarity, text size, and application complexity. Highly dynamic interfaces, very small text, or applications with unusual layouts may produce less precise summaries.
|
||||
|
||||
**Solution:** Check your per-app permissions to ensure the relevant app is included. For applications with small text, increasing display zoom may improve capture quality. If summaries for a specific app are consistently poor, you can exclude it and rely on other context sources instead.
|
||||
|
||||
***
|
||||
|
||||
#### **Auto-complete suggestions feel irrelevant**
|
||||
|
||||
Auto-complete draws on your Neocortex memory context. If you have recently connected sources or just started using Screen Intelligence, the memory may not yet have enough context to generate relevant suggestions.
|
||||
|
||||
**Solution:** Give it time. As more data is indexed, suggestion quality improves. You can also check that your most relevant connected sources are active and that Screen Intelligence is enabled for the applications where you do most of your work.
|
||||
|
||||
#### Trust or risk signals feel inaccurate
|
||||
|
||||
@@ -137,7 +169,7 @@ Trust and risk intelligence is indicative, not authoritative. Signals may lag re
|
||||
|
||||
**Solution:** Use these signals as inputs into your judgment rather than standalone decisions. Over time, as more outcomes accumulate within the same context, signal quality improves.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### A source does not appear in analysis
|
||||
|
||||
@@ -145,7 +177,7 @@ OpenHuman can only analyze sources you have connected and that fall within the s
|
||||
|
||||
**Solution:** Ensure the source is connected in your settings and explicitly referenced or selected in your request. OpenHuman does not automatically include all connected sources by default.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### Integrations do not update as expected
|
||||
|
||||
@@ -153,7 +185,7 @@ Integrations only run when explicitly triggered by your action. OpenHuman does n
|
||||
|
||||
**Solution:** If an export fails, check that the integration is still connected and that permissions are valid. Retrying the action after resolving any permission or availability issues usually succeeds.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### Performance feels slow
|
||||
|
||||
@@ -161,7 +193,7 @@ Response time depends on request complexity, data volume, and current system loa
|
||||
|
||||
**Solution:** Large scopes and long histories require more processing. Narrowing scope and intent improves responsiveness. During early rollout phases, performance may vary as capacity is tuned.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### Revoking access and residual data
|
||||
|
||||
@@ -169,7 +201,7 @@ When access to a source or integration is revoked, OpenHuman immediately stops p
|
||||
|
||||
Previously exported outputs (such as summaries written to Notion or Google Sheets) remain where they were written. OpenHuman does not retain message content after revocation.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
#### When to contact support
|
||||
|
||||
|
||||
@@ -2,28 +2,47 @@
|
||||
|
||||
OpenHuman is built on the OpenClaw architecture and open-sourced under the GNU GPL3 license. This page explains how the major components connect.
|
||||
|
||||
#### The three pillars
|
||||
#### The four pillars
|
||||
|
||||
OpenHuman's architecture rests on three pillars that work together:
|
||||
OpenHuman's architecture rests on four pillars that work together:
|
||||
|
||||
<figure><img src="../.gitbook/assets/V15 — Three Pillars@2x.png" alt=""><figcaption></figcaption></figure>
|
||||
|
||||
**Neocortex** is the memory engine. It ingests data from connected sources, builds knowledge graphs, manages tiered memory, and provides the recall capabilities that power both conscious queries and subconscious processing. Detailed in Neocortex.
|
||||
|
||||
**Local processing** runs AI models directly on your device. Gemma 3 handles chat, vision analysis (for Screen Intelligence), speech-to-text, and text-to-speech. This layer processes raw screen captures, handles Auto-complete suggestions, and manages local encryption. No raw data leaves this layer.
|
||||
|
||||
<figure><img src="/broken/files/9CaipntgroxxUYuIrQXB" alt=""><figcaption></figcaption></figure>
|
||||
|
||||
**Multi-agent orchestration** distributes work across specialized agents rather than relying on a single monolithic model. An orchestrator agent manages routing, personality, and context distribution. Specialist agents handle specific domains: communication analysis, document synthesis, task management, trading. Agents execute in parallel, not sequentially, enabling real-time responsiveness.
|
||||
|
||||
**Privacy-preserving inference** ensures that raw data never leaves the user's device. Data is encrypted on-device with AES-256-GCM. Encryption keys never leave the device. Only compressed metadata and summaries are processed server-side. Detailed in [Privacy & Security](../product/privacy-and-security.md).
|
||||
|
||||
#### How data flows
|
||||
|
||||
<figure><img src="../.gitbook/assets/V16_Data_Flow_Pipeline@2x.png" alt=""><figcaption></figcaption></figure>
|
||||
<figure><img src="/broken/files/7fx1YutriAn2pnRrusN7" alt=""><figcaption></figcaption></figure>
|
||||
|
||||
1. **Ingestion.** Data arrives from connected sources: Telegram, Slack, Gmail, Notion, blockchain wallets, and others. Each source has its own connector that handles authentication and data retrieval.
|
||||
2. **Compression.** Neocortex processes raw data on-device. Semantic deduplication removes noise. Entity resolution links references across sources. Temporal weighting prioritizes recency. The output is a compressed knowledge graph, not raw text.
|
||||
3. **Storage.** The knowledge graph is stored in Neocortex's tiered memory system. Raw data is discarded after compression. Only structured metadata and summaries persist.
|
||||
4. **Conscious processing.** When you make a request, the orchestrator routes it to the appropriate specialist agent(s). Those agents query Neocortex for relevant context, process your request, and return a result.
|
||||
5. **Subconscious processing.** Independent of your requests, the subconscious system triggers periodic memory recalls from Neocortex. These feed into a self-learning loop that surfaces proactive insights, patterns, and recommendations.
|
||||
6. **Output.** Results are presented to you directly or exported to connected tools like Notion and Google Sheets. Only structured, compressed intelligence leaves the device. Raw data never does.
|
||||
1. **Local capture.** Screen Intelligence captures your screen activity. The on-device vision model processes raw screenshots into structured summaries. Raw screenshots are discarded after processing.
|
||||
2. **Ingestion.** Data arrives from connected sources: Telegram, Slack, Gmail, Notion, blockchain wallets, and others. Each source has its own connector that handles authentication and data retrieval. Screen activity summaries from the local layer are also ingested.
|
||||
3. **Compression.** Neocortex processes raw data on-device. Semantic deduplication removes noise. Entity resolution links references across sources. Temporal weighting prioritizes recency. The output is a compressed knowledge graph, not raw text.
|
||||
4. **Storage.** The knowledge graph is stored in Neocortex's tiered memory system. Raw data is discarded after compression. Only structured metadata and summaries persist.
|
||||
5. **Conscious processing.** When you make a request, the orchestrator routes it to the appropriate specialist agent(s). Those agents query Neocortex for relevant context, process your request, and return a result.
|
||||
6. **Subconscious processing.** Independent of your requests, the subconscious system triggers periodic memory recalls from Neocortex. These feed into a self-learning loop that surfaces proactive insights, patterns, and recommendations.
|
||||
7. **Output.** Results are presented to you directly or exported to connected tools like Notion and Google Sheets. Only structured, compressed intelligence leaves the device. Raw data never does.
|
||||
|
||||
#### Local Models
|
||||
|
||||
The desktop app runs the following models on your device hardware:
|
||||
|
||||
**Gemma 3 (chat):** Handles conversational interactions, query processing, and response generation locally. For requests that require deeper reasoning or larger context, processing is routed to the server-side intelligence layer.
|
||||
|
||||
**Gemma 3 (vision):** Processes screen captures from Screen Intelligence. Analyzes visual content to produce structured summaries of what is on screen: application names, visible text, interface elements, and user activity context.
|
||||
|
||||
**Speech-to-text:** Converts voice input to text locally. No audio data is sent to any server.
|
||||
|
||||
**Text-to-speech:** Generates spoken responses locally for voice interaction.
|
||||
|
||||
All local models run on your device's CPU. No GPU is required, though GPU acceleration is used when available. The models are optimized for low memory footprint and fast inference.
|
||||
|
||||
#### Model-agnostic design
|
||||
|
||||
@@ -35,6 +54,8 @@ This is a deliberate architectural choice. AI models are commoditizing. Performa
|
||||
|
||||
OpenHuman is publicly available on GitHub under the GNU GPL3 license.
|
||||
|
||||
**GitHub:** [github.com/tinyhumansai/openhuman](https://github.com/tinyhumansai/openhuman) **Neocortex benchmarks:** [github.com/tinyhumansai/neocortex/tree/main/benchmarks](https://github.com/tinyhumansai/neocortex/tree/main/benchmarks)
|
||||
**GitHub:** [github.com/tinyhumansai/openhuman](https://github.com/tinyhumansai/openhuman)
|
||||
|
||||
**Neocortex benchmarks:** [github.com/tinyhumansai/neocortex/tree/main/benchmarks](https://github.com/tinyhumansai/neocortex/tree/main/benchmarks)
|
||||
|
||||
Contributions, feedback, and issues are welcomed. The project is in early alpha.
|
||||
|
||||
@@ -74,6 +74,20 @@ Neocortex does not store flat text. It builds knowledge graphs: structured repre
|
||||
|
||||
The result: millions of tokens of organizational noise compressed into a structured, queryable knowledge graph that any AI model can reason over in real time.
|
||||
|
||||
#### What Neocortex Indexes
|
||||
|
||||
Neocortex ingests data from multiple source types:
|
||||
|
||||
**Messaging platforms:** Telegram, Slack, Discord, and other connected communication tools. Messages, conversations, group discussions, and channel activity are all compressed into the knowledge graph.
|
||||
|
||||
**Screen activity:** When Screen Intelligence is enabled, the on-device vision model produces structured summaries of your screen captures. These summaries are indexed by Neocortex as context about what you were working on, which applications you used, and what content you were viewing.
|
||||
|
||||
**Productivity tools:** Documents from Notion, data from Google Sheets, and other connected tools contribute entities and relationships to the graph.
|
||||
|
||||
**Blockchain data:** Wallet activity, transaction history, and on-chain events from connected Web3 sources.
|
||||
|
||||
Every source type feeds into the same unified knowledge graph. Cross-source entity resolution ensures that a project mentioned in Slack, visible on your screen in a spreadsheet, and documented in Notion is understood as one entity.
|
||||
|
||||
#### Neocortex and the subconscious
|
||||
|
||||
Neocortex serves as the foundation for OpenHuman's subconscious, going well beyond retrieval.
|
||||
|
||||
@@ -24,6 +24,8 @@ The subconscious does not generate generic AI outputs. It produces context-speci
|
||||
|
||||
It might surface a risk you had not noticed: two team members making contradictory commitments in separate conversations. It might connect a trend in your trading data to a discussion from three weeks ago. It might remind you of a commitment you made that is approaching its deadline, before you think to ask.
|
||||
|
||||
It might notice that the spreadsheet you were reviewing on screen this morning contradicts a decision your team made in Slack last week.
|
||||
|
||||
The outputs are proactive. You do not need to prompt OpenHuman to get value from the subconscious. It works in the background, feeding insights into the system that are available when you interact.
|
||||
|
||||
#### The mirror test
|
||||
|
||||
@@ -2,27 +2,29 @@
|
||||
|
||||
OpenHuman adapts to different coordination and information environments across communication platforms. While the underlying system remains the same, the way value is realized depends on how your tools and conversations are used in each context. This section outlines common use cases to help you understand where OpenHuman fits naturally and where it may offer limited benefit.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Individual Power Users
|
||||
|
||||
Individual power users often participate in many active conversations and channels at once. Important messages are easily missed, and catching up becomes time-consuming and mentally draining.
|
||||
|
||||
In this context, OpenHuman is used to summarize conversations, surface high-signal updates, and provide quick context before responding. Users rely on it to understand what changed, what requires attention, and where follow-ups may be needed without scrolling through entire threads.
|
||||
You ask OpenHuman: "What do I need to know from the last 12 hours?" It returns: your team agreed to move the deadline to Friday in the dev channel, Sarah raised a scope concern in a DM to you that you have not responded to, there is a heated discussion about the API redesign in the architecture group, and two people are waiting for your approval on PRs. That is 247 messages compressed to a 30-second briefing.
|
||||
|
||||
The value here is personal clarity and reduced cognitive load rather than shared coordination.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Traders and Analysts
|
||||
|
||||
Traders and analysts use communication platforms as real-time information sources. Signal quality varies widely, and misinformation, repetition, and scams are common.
|
||||
|
||||
OpenHuman is used to extract claims, track how information resolves over time, and surface signals weighted by historical reliability within context. Risk intelligence helps flag suspicious behavior or repeated failure patterns across groups and channels.
|
||||
OpenHuman extracts claims from conversations and tracks how they resolve over time. When someone in a trading group says "this token is going to 10x by Friday," that becomes a tracked claim with a resolution date.
|
||||
|
||||
Over time, you see who is consistently accurate and who is consistently wrong, within that specific context. Risk intelligence flags behavioral patterns associated with scams: coordinated message timing, impersonation signals, or repeated failure patterns across groups.
|
||||
|
||||
The outcome is better information filtering and reduced exposure to low-quality or malicious signals not automated trading or execution.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## DAO and Web3 Communities
|
||||
|
||||
@@ -32,7 +34,7 @@ OpenHuman is used to preserve decisions, track contributions, and support shared
|
||||
|
||||
The value lies in maintaining continuity and accountability without sacrificing the speed and openness of chat-based coordination.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Community Managers and Moderators
|
||||
|
||||
@@ -42,17 +44,19 @@ OpenHuman supports these roles by surfacing aggregated insights about group heal
|
||||
|
||||
OpenHuman does not replace moderation judgment or enforcement tools. It provides context and early warning signals.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Distributed Teams
|
||||
|
||||
Distributed teams often use messaging platforms for fast coordination, even when formal work happens elsewhere. Decisions and commitments are made in chat and later forgotten or misinterpreted.
|
||||
|
||||
OpenHuman is used to extract actions, preserve decision context, and sync structured outcomes to tools like Notion or Google Sheets when needed. This reduces friction between discussion and execution without forcing teams to abandon their preferred communication platforms.
|
||||
OpenHuman extracts actions and decisions from conversations and can export them to Notion or Google Sheets as structured records. When you ask "What did we commit to this week?", it returns a list with attribution and source context.
|
||||
|
||||
The subconscious might proactively flag a contradiction: "Alex committed to shipping by Friday in #engineering, but Maria said the dependency won't be ready until next Monday in #design."
|
||||
|
||||
The benefit is improved follow-through and clarity without heavy process overhead.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## Service Providers and Contributors
|
||||
|
||||
@@ -62,10 +66,12 @@ OpenHuman helps surface contribution patterns and trust artifacts that reflect c
|
||||
|
||||
The value is portable credibility grounded in behavior rather than claims.
|
||||
|
||||
---
|
||||
***
|
||||
|
||||
## When OpenHuman Is Not a Good Fit
|
||||
|
||||
OpenHuman provides limited value in environments where conversations are purely social, meme-driven, or intentionally ephemeral. It is also not designed for high-frequency trading execution, formal project management, or environments that require strict real-time guarantees.
|
||||
|
||||
Screen Intelligence adds less value in workflows that are primarily non-visual (terminal-only work, audio-only meetings without screen sharing) or in environments where screen content changes very rarely.
|
||||
|
||||
Understanding these boundaries helps users apply OpenHuman where it is strongest.
|
||||
|
||||
Reference in New Issue
Block a user