One place for agent work
Conversations, files, tools, memory, and outputs stay attached to the work.
Agent workspaces for teams
Agents do the work.Teams stay in control.
Deploy CubePlex in your own environment and turn conversations into governed, repeatable work.
Tell CubePlex what you want to get done...
Medium
DeployDocker Compose
ScaleKubernetes + Helm
ConnectYour model providers
KeepYour data and policies
Start with a request.
Finish with the work.
CubePlex gives every team a shared place to ask, act, review, and build on what the agent already knows.
Visible while it happens
See reasoning, tool calls, artifacts, and execution state instead of waiting on a black box.
Boundaries built in
Workspace roles, tool grants, credentials, and approval paths keep action in scope.
Give every agent what the work needs.
Extend capability without turning access into a free-for-all. Each workspace gets the right context, tools, and execution environment.
Work starts where teams are.
Use CubePlex on the web or bring the agent into the IM channels your team already uses.
The workspace assembles context.
Memory, reusable skills, and approved MCP tools shape the run before the agent acts.
Every action stays in scope.
Policies, approvals, sandbox isolation, traces, and audit events make execution governable.
Outputs stay with the work.
Artifacts remain inside the workspace and the same process can run again through automation.
Inputs can vary. Workspace policy remains the boundary.
MemoryRecall team context
SkillsLoad reusable playbooks
MCP toolsUse approved systems
Workspace agentPlan, act, and create
PoliciesScope access and approvals
SandboxIsolate code execution
ArtifactsKeep files and outputs
AutomationSchedule or trigger runs
Traces & auditInspect steps and actions
Start where the team already works
Web / H5
IM
Workspace boundary
Workspace context
Memory, skills, and MCP tools bring approved context into every run.
Governed execution
The workspace agent works through policies, approvals, and an isolated sandbox.
Reviewable outcomes
Artifacts stay with the workspace. Automation can schedule or trigger the next run.
Full visibility
Traces expose each step, while audit events record changes and access.
Useful agents need room to act.
Control should shape the work, not slow it down.
Scope access
Assign tools, models, skills, and data at organization, workspace, or user level.
Approve important actions
Keep sensitive operations behind a clear human decision instead of a hidden default.
Inspect every run
Trace what the agent used, what it changed, and where the result came from.
Your infrastructure.
Your operating boundaries.
Run CubePlex on one host with Docker Compose or deploy to an existing Kubernetes cluster with Helm. Both paths use the same application images.
Read the deployment guideDocker ComposeSingle host
$ git clone https://github.com/cubeplexai/cubeplex.git
$ cd cubeplex/deploy/docker-compose
# Configure .env and production YAML
$ docker compose up -dGive the work somewhere to go.
Set up CubePlex, connect the tools your team already uses, and delegate the first real task.