Inside your environment
Run on your own infrastructure with local storage, your identity configuration, and explicit network access.
DEPLOY ON YOUR TERMS
The same intelligence.
Wherever the mission goes.
Run CyberStitch close to the systems it observes. Keep the interface and core storage local, and connect peers when your network and policies allow.
In the cloud you choose
Deploy a binary or container in your own cloud environment. Configure durable storage, HTTPS, backups, and permitted integrations.
Close to the mission
Keep local visibility useful when connectivity is limited. Plan collection, peer exchange, packages, and upgrades around your network boundary.
A COMPACT CORE
Less supporting infrastructure.
More operational context.
The core Go application embeds its web interface and uses local databases. Optional connectors and local inference components have their own packaging and resource requirements.
Your storage needs depend on collection volume, retained history, output spooling, and the capabilities you enable. Size and monitor the deployment accordingly.
- Core runtime
- A Go executable with the compiled web interface.
- Local storage
- Embedded databases and a persistent application data directory.
- Platforms
- Linux and macOS on amd64 and arm64, subject to release and component compatibility.
- Packaging
- Binaries, tar archives, Linux DEB/RPM, macOS PKG, and Linux container images.
- Extensions
- Connector ZIPs and signed context packages distributed separately.
- AI
- Optional. Supervised local inference or configured external providers, with per-agent model bindings and explicit egress policy.
PLACE INFERENCE DELIBERATELY
Choose where the agents do their work.
The core platform remains useful without AI. When you enable agents, choose provider bindings, models, domain packages, and egress rules together.
Prepare intelligence for the edge.
The supervised llama-server integration uses locally supplied runtime and model files, validated against a checksum manifest. Preposition compatible models and signed context packages, then test memory, latency, and task quality on the deployment hardware.
Use an allowed inference destination.
Supported adapters include OpenAI, Anthropic, and OpenAI-compatible providers. Configure endpoint, credentials, model, and capabilities. These adapters are treated as external destinations even when an endpoint has a local-looking address.
Match the model to the assignment.
Different agents can bind to different configured models and providers. Required tools, structured output, context capacity, and provider readiness affect the available workflows. Generation and embeddings have separate destination checks.
Local-only package or dataset policy can prevent an external inference request. Agent Teams currently coordinate agents within an instance; they are not a cross-site inference mesh. Do not assume an automatic cloud fallback in a disconnected deployment.
PREPARE THE ENVIRONMENT
A deployment is more than an install.
Identity and access
Choose approved administrators, define customer or operator access, and provision credentials with appropriate scope.
Data and connectivity
Decide what stays local, which peers may share state, and which records may leave through output streams.
Recovery and updates
Document backup, restore, disk capacity, and update procedures. Rehearse them under your actual connectivity constraints.
THE WHOLE PICTURE STARTS HERE
Connect your signals.
Know what matters.
Bring your environment. We'll help you see the dependencies,
understand the impact, and find your next move.