Execution
Give every agent a complete, isolated development environment.
Your AI developer needs a safe place to work
CIBOT is the execution and control platform for AI software development. Run coding agents interactively or autonomously inside isolated sandboxes.
Keep access controlled, work auditable, and infrastructure under your control.
>_ Currently in closed development. CIBOT will become open source when the MVP is complete.
>_ Your development stack
Developerhuman input
Issues / Tasksevent input
CIBOT CONTROL PLANE
provision│
▼
▼
ISOLATED SANDBOX
AI coding agentRepositoryToolchainTests / ServicesProjectsPort management
controlled access│
▼
▼
Repos / APIs / Cloud
The platform around your AI developer.
CIBOT separates execution, control, and access into three architectural responsibilities.
Control
Centrally manage agents, sandboxes, projects, identities, policies, and lifecycle.
Access
Let agents use repositories and infrastructure without direct access to sensitive credentials.
>_
HUMANSDeveloper / Reviewer / Platform▼
CIBOT CONTROL PLANE
TasksIdentityPoliciesAuditOrchestration
SandboxAgent
Repo
Runtime
Repo
Runtime
SandboxAgent
Repo
Runtime
Repo
Runtime
SandboxChore
CIBOT ACCESS PLANE
ToolsPoliciesCredential injectionProxy
GitLab
Cloud APIs
Internal systems
One platform. Two ways to develop.
Both modes use the same CIBOT sandbox and isolation boundary. Developers can work with an agent interactively, or let an issue or event trigger autonomous work.
>_ Execution modes
CIBOT SANDBOXCoding Agent
INTERACTIVE
Developer
↓
AI Agent
↓
Iterate together
AUTONOMOUS
Issue / Event
↓
AI Agent
↓
Implement → Test → MR
Interactive development
A developer and agent share the same development environment. The human can inspect or take over the environment at any moment.
Autonomous development
A task triggers a sandbox. The agent works independently, produces a result, and hands it back for review. The developer can at any time connect to the environment and continue with interactive development.
A real computer for every AI developer.
Agents do not work in an abstract chat session. They get a complete computer that you already own.
>_ CIBOT sandbox
VM isolation boundary
AI HARNESSCodex / Claude / …
Development Environment
RepositoryRuntimeDependenciesTestsServicesBrowser / GUI
controlled boundary
▼
▼
Outside world
Repositories can be cloned, applications started, tests executed, browsers launched, and GUI applications exercised inside one isolated environment.
The isolation boundary does not depend on the coding agent that runs inside it. It is a real computer that you can use for any other task you need.
Control multiple AI environments from one place.
The Control Plane connects identity, policy, lifecycle, and configuration to every sandbox. The execution stays isolated and visibility is centralized.
>_ CIBOT Control Plane
Projects
Identities
Policies
Agents
Sandbox lifecycle
Access
Tasks
Audit
Configuration
Sandbox 01Developer + Agent
Sandbox 02Issue #421 Agent
Sandbox 03Chore Agent
The Control Plane knows:
- →who initiated the workload
- →what project it belongs to
- →which agent is running
- →what resources it can access
- →which policies apply
- →what the agent has done
- →when the environment must be created or destroyed
Agents should not carry your secrets.
Keep credentials outside the untrusted execution boundary or properly protected within the sandbox (automatic key rotations, password protection, etc).
>_ Capability flow
Untrusted execution
SANDBOX / AI Agent
No GitLab tokenNo cloud credentialsNo production secrets
capability request
▼
▼
Security boundary
CIBOT ACCESS PLANE
IdentityPolicy evaluationCredential injectionRequest filteringAudit
GitLab
Cloud
AI should not only write code.
Persistent automation can support work across the complete development lifecycle.
>_ Development lifecycle
Issue
→Development
→Merge request
→QA
→Approval
→Merge
→Release
CIBOT Chores
link issueenrich MRupdate statusrequest reviewprocess QA feedbackverify approvalsprepare releasecustom workflows
Automation that responds to the lifecycle.
Chores turn events from development systems into focused actions.
>_ Chores architecture
GitLab
Jira
CI
Schedule
EVENT
▼
▼
CIBOT CHORE
Issue
MR
Pipeline
MR Enrichment
Add issue references, metadata, context, and summaries.
Lifecycle Sync
Move tickets between development, testing, and completion.
Approval Guard
Explain why work cannot move forward.
Feedback Loop
Turn QA feedback into a new agent task.
Everything becomes an auditable event.
Audit data emerges from each system layer and flows into one record.
Prompt → execution → file changes → tests → external actions → result
AUDIT LOG
Sandbox
- Commands
- Files
- Tests
- Agent actions
Control
- Lifecycle
- Policies
- Identity
- Configuration
Access
- Tool calls
- External APIs
- Credentials
- Decisions
Autonomous does not mean unsupervised.
Human review is a defined part of the development workflow.
>_ Supervised workflow
Task→Agent→Implement→Test→Propose change
HUMAN REVIEW
Approve ↓MERGE
Feedback ↻AGENT
Humans can define where approval is required and where agents may continue autonomously.
One execution model. Your infrastructure.
Run the same isolated sandbox on local, cloud, or dedicated compute with sandbox boundary using virtualization.
>_ Deployment model
CIBOT CONTROL PLANE
Developer machine
VMSandbox
Your cloud
VMSandbox
Datacenter
VMSandbox
Local
A persistent developer environment on a workstation.
Cloud
Ephemeral or persistent sandboxes provisioned on cloud compute.
Dedicated infrastructure
Run specialized workloads on your own servers, networks, or hardware.
Bring the coding agent you want.
CIBOT separates the coding agent from the environment around it.
>_ Standard execution layer
CIBOT
standard execution layer
┌──────────┼──────────┐
▼
┌──────────┼──────────┐
▼
CODEX
CLAUDE CODE
FUTURE AGENTS
Switch models or agent harnesses without redesigning your execution, security, and governance architecture.
One platform. Clear value for every team.
Each team uses the same architecture for a different responsibility.
Developers
└──AI development environment
Platform Engineering
└──Reusable execution + capabilities
Security
└──Isolation + controlled credentials
Engineering
└──Autonomous execution + visibility
Built to become open source.
CIBOT is in closed development while we complete the first usable release. We will publish the project under an open-source license when the MVP is ready.
Help us keep building in the open.
CIBOT is part of the Distro Ops open-source and commercial ecosystem. Partners help fund long-term development while the projects remain freely available under their open-source licenses.
Partners
Visibility across the Distro Ops open-source ecosystem. Including project websites and README files + presence on the central Distro Ops partner page.
Partner Slot #2
Partner Slot #3
Give your AI developers a safe place to work.
CIBOT is in closed development. Follow the project as we build toward the open-source MVP.