For day-to-day work

Null Chat

A model-agnostic AI chat for everyone in your company, with the privacy layer on by default.

01 · On by default

Privacy runs in the background.

Every prompt, every attachment, every model call is inspected and tokenized before it leaves. Your people don't toggle anything. They just chat.

02 · Model agnostic

Pick the best model for the job.

GPT-5.1, Claude 4.5 Sonnet, Mistral Large 3, or self-hosted Llama. Admins decide what's allowed. Users just pick from a list.

03 · Nothing leaves

Cleartext never leaves the device.

Re-identification happens in the browser. Server logs, inference traces, training data, none of them ever see a real name or case ID.

Model selection

Switch models mid-thread.

Start a legal review in Claude. Finish the draft in GPT-5.1. Route a sensitive schedule check to your self-hosted Llama. Null maintains the same vault across models, so tokens are stable, references resolve, and audit stays coherent.

Admin-gated allow-list
Workspace admins decide which models your team can choose, per workspace, not per user.
Self-hosted support
Point Null at your private Llama, Mistral or Qwen endpoint. Same privacy layer, zero egress.
Training opt-out enforced
We negotiate no-training addenda with every provider and block non-compliant endpoints at routing.
Active workspaceRouting live
GPT-5.1
OpenAI · EU endpoint · 128k ctx
2 days avg
Claude 4.5 Sonnet
Anthropic · FR · 200k ctx
Mistral Large 3
Mistral · FR · 128k ctx
Llama 3.3 70B
Self-hosted · Frankfurt DC · 64k ctx

Document chat

Drop a contract. Ask anything.

Upload 200 pages of underwriting notes, a patient file, or a claims folder. Null's Document Intelligence engine parses every page, detects sensitive data with structure awareness, and swaps it with stable Vault tokens before the model sees a word.

Native PDF, scanned, images
OCR fallback for scans and photos. Quality scores flagged per page, so reviewers catch low-confidence detections.
Role-aware tokenization
Plaintiff, defendant, claimant, adjuster, roles are preserved across the document so the model reasons correctly.
Stable vault across files
Attach a follow-up document in the same thread and the same person keeps the same token. References survive.
Pipeline64 entities · 138 ms
PDF
claim-2047-hoefer.pdf
14 pages · 842 KB
Tokenized
DOC
policy-HV-2020-33891.docx
8 pages · 214 KB
Tokenized
IMG
damage-report-scan.tiff
3 pages · 2.1 MB · OCR
Scanning
PERSON_A·S7FT9CD → Hermann Höfer · confidence 0.98 (NER+struct)
ORG_A·44LT3R → Hanseatic Versicherung AG · confidence 0.99
ADDR_A·9MPLR2 → Fährkanalstraße 12, Hamburg · confidence 0.96
POLICY_A·33891 → HV-2020-33891 · confidence 0.94 (rule)
DATE_X·81QKP2 → 2026-03-14 · confidence 0.99
AMOUNT_A·18420 → €18,420 · confidence 0.91

Prompt library

Company playbooks, on tap.

Your compliance team writes the prompts once. Everyone runs them with one click, with the right variables, the right model, and the right token scope. No more copy-pasted prompts drifting across Slack.

Version-controlled
Every prompt has an owner, a changelog, and approval gates for high-risk use cases.
Bound to workspaces
Share prompts across a team, a department, or lock them to a single regulated workspace.
Run-through audit
Every invocation logs inputs, tokens, output, and reveals, so risk can reconstruct decisions.
Workspace · Schadensfälle Q212 prompts
Claims · v2.124 runs today
Summarize claim and list disputes
Attach a claim file. Outputs a structured summary plus a list of contested line items with policy section references.
Underwriting · v1.48 runs today
Risk scan: industrial property
Attach inspection report. Outputs a risk-profile with cited clauses and recommended premium adjustment band.
Compliance · v3.0 · approved2 runs today
KYC red-flag review
Attach customer file. Outputs a red-flag checklist with citations. Approval-gated; every run creates an audit record.

Workspaces

Shape it to your org.

Every team, department, or case folder gets its own workspace, with its own models, prompts, knowledge base, and audit trail.

Isolated by design

Each workspace has its own vault, its own allow-list, and its own audit log. Cross-contamination between cases is architecturally impossible.

Invite by role

SCIM-provisioned groups. SSO-gated access. Role-based prompts, knowledge, and reveal scopes, no one toggles a setting they shouldn't.

Knowledge base

Upload policy libraries, benefit schedules, pricing decks. Employees can ground every answer in your own sources, tokenized if they contain sensitive data.

Continuous audit

Every prompt, every model hop, every reveal, written to a tamper-evident log. Export for internal audit, regulator, or court order in one click.

Compliance, built in

Designed for the auditor, not just the user.

Null Chat was designed from the governance side first. DPA, ROPA templates, Art. 32 “state of the art” pseudonymization, and an EU AI Act risk-classification worksheet , all shipped with the product.

Pseudonymization
GDPR Art. 32
State of the art
Hosting
Frankfurt · DE
EU-only routing
Certifications
Attestations available
ISO 27001 · 42001
EU AI Act
Risk worksheet included
Limited risk · ready

See Null Chat on your own document.

Bring a real claim file, contract, or medical note. We'll load it into a sandbox workspace with you. You keep the vault. Nothing leaves the room.