How feder8d keeps your data yours.
The same modern AI your team wants — chat, search, assistants — running in a private space your company controls. Below is the plain version of how that stays safe, and the detail your security team will check.
What sees your data
- Your documents
- Stay in your space
- Open-weight models
- Answer, keep nothing
- Public AI services
- Never in the path
- Model training
- Your data is never used
Everything happens inside your space.
A question goes in, an answer comes back, and the whole round trip stays within the boundary you control. The model does its work and keeps nothing — your data never becomes part of a model and never leaves your space.
Inside the boundary
- Your documents
- Your AI assistant
- Your encryption keys
- Your audit record
Sealed — your data stays in here.
Outside the boundary
- Public AI services
- Never see it.
- Outbound network calls
- Off unless you enable them.
Nothing crosses without an explicit rule you wrote.
Separated at every layer.
Your company doesn't just get a login — it gets its own space at every level. Five separations, so one mistake can never expose your data to anyone else.
There is no shared pool anywhere. Your data is never mixed with another customer's — not even with ours.
| Database | Your records live in a database of their own — never a shared table. |
|---|---|
| Search index | Your documents are indexed in a space only your company can query. |
| Cache | Working memory is scoped to you. No spill-over between customers. |
| Encryption keys | Your data is encrypted with keys that are yours alone. |
| Network boundary | Locked down by default — nothing leaves without an explicit rule. |
What we don't compromise on.
The promises behind the product — the ones your security team will check.
No public AI in the path
Your data never reaches ChatGPT, OpenAI, or any outside AI service — unless you deliberately turn it on for a specific use. Off by default.
Separated, not just labelled
Each company runs in its own space. A mistake in our software still can't expose your data to another customer — there's no shared pool to leak.
Open models, never closed black boxes
We run open-weight models with known, audited licences — so you always know what's processing your data. No mystery foundation models.
A record you control
Every action is written to a log that lives with you. Your auditors and DPO get straight answers, not 'trust us'.
Open models you can trust.
We run open-weight models with real, published licences — so you always know what's handling your data. No closed, "trust us" foundation models in the path.
See every model we ship| Model | Used for | Licence |
|---|---|---|
| Qwen 2.5 | General chat & retrieval | Apache-2.0 |
| Mistral Small | Longer, nuanced work | Apache-2.0 |
| Llama 3.3 | Heavier reasoning (opt-in) | Llama Community |
Your region, your compliance.
Your data stays in the jurisdiction you choose, encrypted in transit and at rest — because the software runs where you put it.
- Data residency in your jurisdiction
- Self-service data export from the console
- 72-hour breach-notification process wired to your contact
- No public AI in the prompt path — optional BYO provider only on explicit opt-in
| Control | Status |
|---|---|
| GDPR | compliant |
| POPIA | compliant |
| SOC 2 Type II | observation in progress |
| Encryption at rest | enabled |
| Encryption in transit | enabled |
Launched 2026-06-01. Posture is generated from our published compliance configuration, not written by hand.
Source available to your auditors on request.
feder8d is licensed under the AGPL-3.0. We provide the corresponding source to customers, users, and their auditors on request (under NDA where appropriate) — so your security team can review exactly how we handle your data, rather than taking the privacy guarantees on faith.
Keep your data where it belongs.
Give your team private AI today — free for 14 days. Your sensitive data never reaches public AI.