How Skynet Works
1. Overview
Skynet is an AI workspace that brings multiple frontier models, autonomous agents, and your live business context together in one place. This page explains how the pieces fit together: from how requests are routed, to how agents work in parallel, to how your organization’s knowledge is built and protected.
If you’re new here, read top to bottom. If you’re looking for something specific, use the table of contents to jump to the section you need.
2. Smart Model Routing
Every request you make is handled by the model best suited for the job, automatically.
- Fast model
- Balanced model
- Reasoning model
- Specialist model
Skynet works across multiple frontier models. When you submit a prompt, the routing system evaluates the task type, the quality requirements, and the cost implications, then selects the model that fits. A quick draft might go to a faster, lighter model. A complex analysis might go to a more capable one. You don’t have to think about which model to use.
Manual override. You can pin a specific model to a conversation or switch models mid-thread if you prefer. The choice is always yours.
Enterprise controls. Admins can set policies that map specific models to specific data types or projects. Every model selection is recorded in audit logs, so you always know which model handled which request and why.
3. Parallel Sub-Agent Forking
When you give Skynet a large objective, it doesn’t tackle everything in one thread. It forks.
- Sub-agent inherits the skill chain
- Sub-agent inherits the skill chain
- Sub-agent inherits the skill chain
A parent agent breaks the objective into components and spawns sub-agents to handle each one. Every sub-agent inherits the full skill chain, meaning the tools, context, and capabilities of its parent, so it can work independently without losing capability.
Each sub-agent works on its piece in parallel. When they’re done, the results merge back into one coherent output. You see a single, unified result, not a pile of fragments.
This is what makes Skynet scalable. Because sub-agents share memory and context, the system can run thousands of parallel tasks without losing coherence or duplicating effort.
Here’s the full flow, from the moment you state an objective to the moment you review the result.
- Give your objective Tell Skynet what you want, in plain language. It can be as simple or as complex as you need.
- Break it into parallelizable components The system analyzes the objective and identifies which parts can be worked on simultaneously and which need to happen in sequence.
- Fork sub-agents The parent agent spawns sub-agents for each component. Each inherits the skill chain and relevant context.
- Each sub-agent works independently Sub-agents execute their assigned pieces in parallel, drawing on tools, memory, and live data as needed.
- Merge results The parent agent collects and synthesizes the sub-agents outputs into a single coherent result.
- Review You review the merged output, give feedback, and iterate. You are always in control of the final result.
4. Real-Time Context Grounding
Skynet doesn’t work in a vacuum. It connects to your live data sources, including email, chat, meetings, documents, and Jira, so its responses are grounded in what’s actually happening in your work.
This context is maintained continuously in the background. You don’t have to paste in context or re-explain your situation every time. When you ask a question, Skynet already knows what project you’re working on, what was discussed in yesterday’s meeting, and what’s open in your task tracker.
The result: answers that reflect current reality, not generic best guesses.
5. Unified Memory Architecture
Skynet’s Unified Memory Architecture gives every agent, model, and workflow a single, persistent memory that spans conversations, tasks, and connected data sources. Instead of starting fresh each time, Skynet remembers, and gets smarter with every interaction. Think of it as a company brain: a central repository where your organization builds and acts on collective knowledge, automatically.
Three Memory Layers
- Individual Memory Private by default Your tasks, preferences and interactions. Nobody sees them unless you share.
- Team Memory Shared with your team Meeting notes, project discussions and team decisions everyone can draw from.
- Company Brain Organization-wide Patterns, history and context that span the whole company.
The Company Brain auto-builds from daily interactions (meetings, chats, emails, connected tools) and does three things: identifies incoming requests and spins up the right agents automatically, recognizes repeating patterns and suggests turning them into skills, and pulls historical context before meetings so relevant background is ready before you walk in.
How Does It Work?
Every piece of information passes through four stages.
- Observe The system watches activity across your workspace and filters for relevance.
- Encode Raw observations become structured, searchable memory.
- Store Memories are persisted with metadata: type, source, timestamp.
- Consolidate Redundant entries merge, contradictions resolve, low-value memories fade.
When you make a request, the system builds context in real time.
- Parse your intent
- Retrieve relevant memories Across all three layers, using semantic search
- Resolve conflicts What you said five minutes ago overrides what the system guessed last week
- Assemble the context
- Act with the full picture
The result: you say “fix the formatting issue from last time” and the system already knows which issue, which file, and which fix worked. No clarification needed.
Enterprise Controls and Why It Matters
Individual memory stays private by default. Admins define what enters shared layers. Projects can be locked down. Your data never trains external models, and every memory entry is traceable.
Most AI tools are powerful in the moment and amnesiac over time. The Unified Memory Architecture changes that. A system that remembers learns, adapts, and compounds value: every interaction makes the next one better. That’s the foundation of genuine autonomy: a system that can’t remember can’t learn, and a system that can’t learn can’t be trusted to act on its own.
6. Skills as Reusable Automation
A skill is a captured work pattern: a repeatable sequence of steps that produced a useful result. Skills turn one-off work into automation you can reuse.
- The Copilot watches how you work
- It spots a pattern Something you do repeatedly
- It proposes a skill
- You review and approve
- That workflow is available as a single action
On Teams and Enterprise plans, approved skills can be shared across the team, so everyone benefits from the automation you’ve built.
7. The Copilot
The Copilot is your always-on workspace partner. It doesn’t wait to be asked. It watches, learns, and proposes.
- Proposes, you approve. The Copilot suggests actions, skills, and agent assignments. You decide what runs.
- Learns your preferences. Over time, it adapts to how you work: your style, your priorities, your routines.
- Spots tasks and assigns agents. When the Copilot sees something that needs doing, it can recommend or dispatch an agent to handle it.
- Works while you’re away. The Copilot can continue working in the background and deliver a morning brief summarizing what happened, what’s ready for your review, and what needs your attention.
- Voice-enabled. You can talk to the Copilot directly, and it’s aware of all activity across your workspace.
8. Enterprise Security and Deployment
Skynet is built for organizations that need control over where their data lives and how it’s governed.
Deployment options. Choose what fits your requirements: cloud, on-prem, your own data center, or your own cloud environment.
Compliance. Skynet meets SOC 2, ISO, and GDPR standards.
Data protection. Your data never trains external models. What’s yours stays yours.
Admin controls. Administrators can manage shared memory policies, control what enters the Company Brain, and set boundaries on what’s shared across teams.
Privacy by default. Individual work is private by default. Projects can be locked. Sharing is intentional, not automatic.
9. Putting It All Together
Skynet brings together several components that work as one system:
- Smart Model Routing sends each request to the best model for the job.
- Parallel Sub-Agent Forking lets the system tackle large objectives by breaking them into parallel tasks and merging the results.
- Real-Time Context Grounding keeps every response tied to your live business data.
- Unified Memory builds knowledge across individual, team, and company layers, and keeps it private where it should be.
- Skills turn repeated work patterns into reusable automation.
- The Copilot ties it all together as an always-on partner that proposes, learns, and works alongside you.
- Enterprise Security ensures all of this happens on your terms: your deployment, your policies, your data.
The core design principle is simple: Skynet augments human capability. It doesn’t replace it. The system proposes, you approve. The system works, you review. You’re always in control.