Bring your own model.
Governed by Obiguard.
Obiguard Desktop is a governed AI workspace and local AI proxy. Your people and their tools use models in the cloud, on your network or on their own machine, and every prompt is checked and recorded by Obiguard.
Every prompt is checked by Obiguard before it reaches the model, and recorded for your organization.
Your model. Your infrastructure.
Desktop doesn't ship a chat model. You choose one, and Obiguard checks what goes in and out.
Open models, run locally or privately
Run them with
Or your private cloud deployment
Product names and logos belong to their respective owners and are shown to indicate compatibility only.
Every prompt is checked first.
Obiguard's gateway checks the prompt before any model sees it, and records the result.
You write a prompt
In Desktop, or in a tool pointed at its local proxy.
Obiguard checks it
Personal data is redacted, injection and risky content are caught, and the endpoint is verified.
It reaches your model
Hosted models are called by the gateway. Models on your network are called directly from your device.
The reply is checked and recorded
Then shown, with the decision and a trace saved to your audit trail.
One hop. The gateway checks the prompt, calls the model and records the trace, so the check cannot be skipped.
The prompt goes straight from your device to your endpoint, so the model never has to be reachable from the internet. Your device must be able to reach it, on the office network or a VPN.
One client. Four layers.
Chat with the model you choose, under your policy.
A familiar workspace for hosted models and endpoints on your network. Every prompt and reply is checked by Obiguard before it is used.
- Personal and confidential data redacted before it reaches the model
- Approved-endpoint allowlist set by your admins
- A record of every conversation for your audit trail
Your coding tools get the same checks.
Point Claude Code, Codex CLI or any SDK at Desktop. Requests are checked and recorded exactly like chat, and the only thing that changes in the tool is its base URL.
- An OpenAI- and Anthropic-compatible endpoint on this machine
- The same policies and the same audit trail
- Blocked requests return the errors your SDK already handles
Know how AI is used. Tell staff exactly what you see.
When your organization turns it on, Desktop records AI usage and device posture against your frameworks. Staff see a plain list of what is recorded and who can see it.
- Read-only device checks, never file contents
- A visible indicator when monitoring is on
- A plain list of what is recorded, and who can see it
Agents and tools that inherit your governance.
Developers will publish agentic workflows that run under the same policy: declared permissions, default-deny network and file access, and admin approval to install.
- Sandboxed, signed apps
- Permission broker and egress allowlist
- Org catalogue with admin approval
The whole stack, inside your walls.
Run Obiguard's control plane, gateway and guardian on your own infrastructure. Your prompts, replies and audit trails stay where you put them, and your people use the same Desktop app they would use in the cloud.
Nothing in the request path leaves your network. Obiguard's cloud is not part of it.
Data stays inside your boundary
Prompts, replies, traces and audit logs are processed and stored on infrastructure you control, in the region or facility you choose.
Models of your choice
Run open-weight models on your own GPUs, in your own cloud tenant, or both. No outside vendor sits between your staff and your models.
Policies you own
Write guardrails, control packs and model allow-lists yourself, and review changes like any other change to production.
Evidence on demand
Exportable audit trails mapped to frameworks such as NIST AI RMF and ISO/IEC 42001, with export to your SIEM.
SIEM export plannedYour identity provider
Single sign-on, SCIM provisioning and role-based access through the identity system you already run.
PlannedRolled out like any enterprise app
Managed deployment through your MDM, signed builds, and updates that follow your change windows.
MDM deployment plannedStart with a scoped pilot.
Enterprise pilots come with a dedicated solutions engineer, security review documentation and a plan sized to your environment, so your team can see it running on your own infrastructure before you commit.
What gets recorded, and who sees it.
A tool that checks every prompt has to be plain about what it keeps.
Prompts and replies
Checked by the gateway and recorded as a trace in your organization’s audit trail.
Decisions and usage
Which policy applied, what was redacted or blocked, which model answered, and how much it used.
Device checks, only if enabled
Read-only settings such as disk encryption. Never files, browsing data or personal data.
You can see it too
Desktop tells staff what is recorded and who can see it, with a visible indicator when monitoring is on.
Retention and access are set by your organization's admins. With Enterprise, all of it stays on your infrastructure.
Common questions.
Bring your own model.
Governed by Obiguard.
Download Obiguard Desktop and sign in with your work email.