Lacuna

Redact PII. Prove it's gone.

Detection runs on our own servers — your pages never go to a third-party AI. Review every span, then commit: the text is removed from the PDF text layer, not just covered, and a hash-chained tamper-evident audit log proves what you removed. For solo attorneys, FOIA officers, and journalists.

Start Solo — $5/mo

Try the free demo

Free demo. No credit card. Files deleted from primary storage on completion.

Why we exist

The privacy gap

AI redaction tools

The new wave finds PII by shipping every page to a third-party AI. Your client's documents leave your control to be read by someone else's model.

The proof gap

Black-box redactors

They cover the text and hand back a file. Nothing lets you independently confirm it's actually gone — or that the log wasn't quietly edited after the fact.

Lacuna

Provable. And private.

Detection runs on our own servers — never a third-party AI. Every job carries a hash-chained, tamper-evident audit log, and you can re-check the output yourself with pdftotext.

Who it's for

Built for the people who redact for a living.

Bulk auto-detection, span-by-span review, and a hash-chained tamper-evident audit log — applied to the documents your work depends on.

Law firms

Redact discovery productions in bulk and keep a defensible record of every call.

See how it works

FOIA & public records

Clear the request queue — auto-detect PII, apply your exemptions, and release with an audit trail.

See how it works

Journalists

Redact documents for publication so the text is gone, not just hidden under a black box.

See how it works

How it works

Three steps. About fifteen seconds.

Bulk PDF redaction with a hash-chained tamper-evident audit log — sensitive spans detected automatically and removed from the PDF text layer.

Upload

Drop a PDF. Up to 25 pages on Free. Files are encrypted in transit and at rest.

Review

Toggle any auto-detected span, add your own terms, jump-to-page before commit.

Download

Text is removed from the PDF object stream. Verify with pdftotext or pdfinfo.

Detection engines

AI detection, on our infrastructure.

Every job runs a deterministic pattern pass plus an on-server AI pass, and combines the results. Both run on Lacuna-operated infrastructure — your documents are not transmitted to any third-party AI service. Orgs that want pattern-only detection can opt out in one toggle.

AI detection

Recommended

Pattern engine + on-server AI model

Default. Every page gets a deterministic pattern pass plus a compact AI model trained for PII detection — and the combined result. Highest recall: catches names + addresses that pattern-matching alone misses.

Pattern pass + AI pass on every page
Bbox-deduped union of both results
Same audit trail regardless of which pass surfaced a span

Data flow: Both passes run on Lacuna-operated infrastructure. Your documents are not transmitted to any third-party AI service, and nothing is trained on them. The model's weights ship pinned inside our worker image.
Retention: File deleted from primary storage on completion; encrypted backups purged within 30 days. No AI-provider retention applies — there is no AI provider in the path.
Available on: Free, Solo, Team, API, Enterprise

Pattern-only

Deterministic rules

Customer opt-out for orgs that want fully deterministic, rule-auditable detection — no model inference of any kind.

Pattern-driven, deterministic
Runs entirely on our infrastructure
Reproducible: same input, same spans

Data flow: Not transmitted to any third-party detection service. Stays on Lacuna infrastructure.
Retention: File deleted from primary storage on completion. Database backup snapshots roll off within the sub-processor recovery window (typically ~7 days).
Available on: Any tier — toggle in Settings → Organization

Trade-off: lower recall on free-form text (names without context, non-Western names, ALL-CAPS headers). The demo above runs both passes by default for this reason.

Pricing

Honest tiers, hard caps.

Hard page caps per tier so one batch upload doesn't blow up everyone else's margins. Overage is metered, never surprise-billed.

Solo

$5/mo

Indie / single attorney

200 pages / month
$0.05 per page overage
Review window: 24 hours
7-day retention
Tamper-evident audit log
AI detection on — runs on our infrastructure, opt-out in org settings

Start Solo

Team

Recommended

$15/seat/mo

Small firms (5–30 attorneys)

1,000 pages / seat, pooled
$0.03 per page overage
Batch upload — up to 50 PDFs per submit
Matters — group jobs by case for discovery / FOIA workflows
Review window: 7 days, share with co-counsel
AI detection on — runs on our infrastructure, opt-out in org settings
30-day retention
Invite teammates — share jobs and audit log
Owner / member roles + tamper-evident audit log

Start Team

API

$25/mo

Programmatic / integrators

5,000 pages / month
REST API + webhooks
API keys — issue, rotate, revoke
Async batch submit + status polling
AI detection on — runs on our infrastructure
Tamper-evident audit log

Free

Trust demo$0

5 redactions / day, ≤25 pages, watermarked. 1-hour retention. No card.

Try the demo

Enterprise

Contact$1k+/mo

Unlimited pages, BYO-LLM (your own Anthropic / Azure / on-prem key), on-prem / VPC deploy, DPA available + BAA on request, dedicated support + SLA.

Talk to sales

The honest part

What you're actually paying for.

Every job runs a deterministic pattern pass plus an on-server AI pass by default — combining both is the difference between catching a non-Western name on a resume and missing it. The AI model runs on our own workers, so there is no per-token bill to pass through to you: per-page all-in cost (detection compute + storage + workers + queue + sandbox + audit log + retention + support overhead) is roughly $0.03. Solo / Team page allowances are sized so a typical workload comes in well under the overage line at $0.03/page beyond. Daily job caps bound the per-IP cost on the free hero demo. Orgs that want fully deterministic detection can opt out — opted-out orgs route through the pattern engine only and we charge the same; what vanishes is the recall lift, not the price. Enterprise can additionally bring its own LLM key (Anthropic, Azure OpenAI, or on-prem) as a second-opinion layer — you pay that model cost directly and we charge for the platform, not the inference.

Clinical research or regulatory submissions? Talk to sales →

Security & compliance

Concrete, not vibes.

Trust signals you can verify, not badges. The audit-log chain ships with a sample you can validate before signup.

Audit log · live chain

Each event hashes the one before it. Go on — click an event to tamper with it and watch the chain break.

● verified

TLS 1.3 only

Modern ciphers. HSTS preload. No legacy fallback.

AES-256 at rest

Provider-managed encryption on Cloudflare R2 and Supabase Postgres.

Deleted from primary storage on completion

Source PDF dropped from R2 at terminal state; Postgres PITR backups roll off within sub-processor windows.

Hash-chained audit log

SHA-256 chain over every event. Verify the chain end-to-end with openssl.

Sub-processor list

Cloudflare, Supabase, Fly, Stripe, Resend, Clerk, Sentry — what each one sees. No AI provider on the list: detection runs on our own workers.

DPA available

Standard Contractual Clauses Module 2/3, signable on Team and above.

API

Drop-in REST. Webhooks on completion.

One endpoint to redact. One webhook to know it shipped. Available on the API tier and Enterprise.

POST /v1/redact

webhook · job.completed

// HMAC signature in X-Lacuna-Signature.
{
  "event": "job.completed",
  "data": {
    "job_id": "a8f3c1b2-...",
    "status": "completed",
    "output": { "download_url": "...", "expires_at": "..." },
    "redaction": {
      "engine": "parallel",
      "spans_applied": 47,
      "entity_counts": { "PERSON": 12, "EMAIL": 8, ... }
    }
  }
}

Join the API waitlist·stable · versioned at /v1

FAQ

The questions buyers actually ask.

Conversion-blocking objections first. If yours isn't here, mail [email protected] — we'll add it.

No. Detection runs on Lacuna-operated infrastructure: a deterministic pattern pass plus a compact AI model purpose-built for PII detection, hosted on our own workers with pinned weights. Your documents are not transmitted to any third-party AI service, and nothing is trained on them — there is no AI provider in the path to retain your text. Orgs that prefer fully deterministic detection can switch off the AI pass in Settings; opted-out jobs run the pattern engine only.