Security
Your environment is the boundary.
Nablo runs in your environment and reads only what you select. Nothing leaves unless you set up the connection.
Illustration with sample data. In the product, attestation is re-checked every 5 minutes.
It runs where your data lives. Nablo runs in your own environment: your servers, your cloud, wherever you choose. It reads only the sources you select, and it never writes to them.
You choose where the model runs. By default, an open-weight model runs in your environment. When a chat needs a bigger model, you can switch it to confidential inference on attested hardware. Nablo checks the attestation first and verifies the signature on every response.
Nothing else leaves. There is no telemetry and no fallback to an outside AI service. Anything outside your environment works only through a connection you set up.
By default
What leaves your environment.
- Stays inside
- Only when you choose
- None
- Your documents and databases
- Stay where they are. Nablo reads only the sources you select and never writes to them.
- The model
- By default, it runs in your environment. You can choose confidential inference for a chat that needs a bigger model; that chat then runs outside your environment, on attested hardware.
- Memory
- Stays in your workspace, scoped to your access rules, and can be undone.
- Learning
- Practice tasks and traces from real runs stay inside your environment.
- Telemetry
- None.
- Outside AI services
- None by default. Confidential inference, or any other outside service, runs only if you choose it.
Confidential inference
Bigger models, on hardware you can verify.
Some work needs a bigger model than your environment can run. For that chat, Nablo can use confidential inference: a model on GPUs that prove what they are running.
The GPU server reports a measurement of the hardware and software it runs.
Nablo checks that the measurement comes from genuine, protected hardware before it sends anything. If the check fails, nothing is sent.
Every response comes back signed. Nablo checks the signature before it uses the answer.
Attestation is checked again every 5 minutes.
Access
Access follows your rules.
Each person sees only what their own permissions allow. Copying information into Nablo never widens who can see it, and memory and results follow the same rules.
For your security team
Questions we hear first.
Not by default. Nablo, its model and its memory run in your environment. Data leaves only through a connection you set up, such as confidential inference for a chat.
No. There is no telemetry.
No. It reads the sources you select and never writes to them.
An open-weight model in your environment by default. For a chat that needs more, a bigger model through confidential inference.
It builds practice tasks from your data and improves itself on them, inside your environment. Traces from real runs stay inside too.
It is rolled back. A new version replaces the current one only if it scores higher on tasks it never trained on.
Start a conversation
Start with one piece of work.
Tell us the work and the data it needs. We will show you how Nablo would learn it inside your environment.
isaac@nablo.io