Core Concepts Hubify is two things: a workspace platform and an intelligence network. Everything below connects these two layers.
Workspaces A workspace is your personal AI OS instance at yourname.hubify.com. Each workspace is:
A Fly.io Machine running OpenClaw gateway + your dashboard (Next.js)
A persistent volume at /data for workspace files, memory, and tasks
A Convex hub for real-time intelligence sync
A subdomain on *.hubify.com with automatic HTTPS
A live terminal at yourname.hubify.com/terminal
Workspace Modes Mode Description Target Guided (default)Reserved structure on, template active, skills pre-installed New users Power Reserved structure on, user customizes freely Builders Open Reserved structure off, raw OpenClaw Advanced users
Multi-Workspace One account supports multiple workspaces for different purposes:
houston.hubify.com -> personal OS (MyOS) bamf.hubify.com -> agency operations (Founder OS) research.hubify.com -> deep research (Research OS) client1.hubify.com -> client delivery (Client OS)
Each workspace is isolated (separate Fly Machine, separate Convex hub) but can share vault and intelligence if configured.
Templates Templates are the viral mechanism. A template is not just a UI skin — it is a complete agent OS:
Dashboard UI — the Next.js web appPre-installed skills — from the Hubify registrySOUL.md — agent identity pre-configured for the use caseAgent roles — orchestrator, researcher, builder, etc.Squad configurations — multi-agent teams with coordination logicAutomations — cron jobs, heartbeat checklists, proactive behaviors
Anyone can fork a template, customize it, and publish it to the template gallery. Template metrics track installs, remixes, and ratings.
The Reserved Structure Every workspace uses a standard file layout that any agent on any platform can navigate immediately.
Reserved Files File Purpose HUB.yamlThe manifest — links everything, model routing, skills, agents AGENTS.mdAgent instructions (cross-platform) SOUL.mdIdentity and persona USER.mdUser profile MEMORY.mdLong-term curated memory HEARTBEAT.mdProactive checklist TASKS.mdTask index (read-only, auto-generated from Convex) LEARNINGS.mdLearnings index (auto-updated by agents)
Reserved Directories workspace/ .hub/ # Runtime state (gitignored, synced via Convex) memory/ # Episodic logs (YYYY-MM-DD.md) skills/ # Installed skills knowledge/ # Curated evergreen knowledge learnings/ # Extracted structured learnings projects/ # User space -- anything goes vault/ # Encrypted credential cache (gitignored)
TASKS.md is read-only. Agents write tasks to the Convex API. The file regenerates on every sync. This prevents the “tasks in markdown that nothing reads” problem.
Skills Skills are the core unit of Hubify. They are executable knowledge packaged as living .hub files that agents discover, install, execute, and improve through use.
5,892+ skills are already in the registry.What Makes a Skill Component Description Prompt The core instruction set Metadata Category, platforms, use cases Trust Metrics Confidence scores from real agent executions Lineage Fork history and evolution path
Trust Metrics Every skill carries trust metrics derived from real execution data — not download counts:
Confidence Score: 0-100% Execution Count Total times used by agents Success Rate Successful completions Agent Diversity Unique agents using it Evolution Count Times improved through learning loop
The Learning Loop
Discover
Agent finds a skill matching their task
Execute
Agent uses the skill
Report
Agent reports success, failure, or improvements
Aggregate
System combines learnings from all agents
Evolve
Skills automatically improve — canary tested, then promoted
5-Gate Trust Gateway Every skill passes through security validation:
Schema validation
Provenance verification
Content security scan (prompt injection, malware, data exfiltration)
Reputation check
E2B sandbox testing
Souls Souls are agent personality templates stored as SOUL.md files. They define who an agent is: its identity, communication style, principles, and constraints.
name : pragmatic-engineer persona : Senior engineer focused on maintainability principles : - Prefer simple solutions over clever ones - Consider long-term maintenance costs - Document decisions, not just code capabilities : - Code review with architectural context - Refactoring for readability - Technical debt assessment
Templates include soul defaults. Users customize their soul and can publish variants to the souls registry. Squads have multiple souls — each agent in the team gets its own identity.
Squads Squads are pre-configured multi-agent teams that ship inside templates. Each squad defines an orchestrator and specialist agents with model routing:
squad : name : "Research Team" orchestrator : soul : "orchestrator.soul.md" model : "anthropic/claude-sonnet-4-6" specialists : - id : "researcher" soul : "researcher.soul.md" model : "kimi/k2.5" spawn_on : [ "research-task" ] - id : "writer" soul : "writer.soul.md" model : "anthropic/claude-sonnet-4-6" spawn_on : [ "write-task" ] automations : - cron : "0 9 * * 1" task : "Weekly research digest"
Squads run on dedicated Fly.io compute. Users can customize squad composition, swap models, and add agents. Community members can publish squad configurations as standalone components.
The Intelligence Layer What makes Hubify more than “just a hosted OpenClaw” is the intelligence layer connecting all workspaces.
Four Levels Level Scope What It Does Agent Individual Profile, personal memory, platform, skills, activity Workspace Cross-platform Shared memory, tasks, learnings across local + cloud Org Private collective All workspaces in a company share memory and vault Global The Singularity Opt-in anonymized learnings from all workspaces
Memory Architecture Each workspace has three memory types, all Convex-backed and real-time:
Episodic — time-based logs (memory/YYYY-MM-DD.md): what happened, whenSemantic — vector-indexed knowledge: searchable across all historyProcedural — skills and how-to knowledge: linked to skills registry
All memory is accessible across local and cloud agents in the same workspace.
Local-to-Cloud Sync Your local OpenClaw and cloud workspace share one brain via Convex real-time:
Synced Not Synced (local only) Memory (episodic + semantic) Raw files (your codebase) Tasks and projects Local shell environment Installed skills Platform-specific configs Learnings Vault access Agent session summaries
HUB.yaml The manifest file that ties everything together. Every connected agent reads it on startup.
hub : version : "2.0" id : "hub_abc123" name : "Houston's OS" template : "myos" intelligence : global : opt-in agents : - id : "agent_cloud" platform : "hubify-cloud" role : "orchestrator" - id : "agent_local" platform : "openclaw-local" role : "primary" models : default : "anthropic/claude-sonnet-4-6" routing : orchestrator : "anthropic/claude-sonnet-4-6" researcher : "kimi/k2.5" automation : "google/gemini-flash" skills : installed : - { id : "github" , version : "2.0.1" } - { id : "strava" , version : "1.2.3" } privacy : contribute_to_global : false vault_shared_across_workspaces : true mode : "guided"
Skills and the reserved structure work across any AI coding agent:
Platform Status Claude Code Full support Cursor Full support Windsurf Full support OpenClaw Full support Custom agents SDK available
The same skill definition works everywhere. Hubify abstracts platform-specific differences.
Security
5-Gate Trust Gateway for all skills entering the registryAES-256 vault — Hubify never stores decryption keysEd25519 agent identity — cryptographic agent verificationScoped tokens — agents receive short-lived credentials, never raw keysWorkspace isolation — each workspace is a separate Fly.io MachineAudit logging for all vault access
Next Steps
Skills Deep Dive Skill structure, evolution, and the learning loop
Workspaces Workspace provisioning and management
Squads Multi-agent team configuration
CLI Reference All 40+ commands
