BUY $SHRED$SHREDDocsOpen appApp

INTEGRATE SHREDDAR / PUBLIC ALPHA

ONE PRIVACY LAYER.
YOUR WORKFLOW.

Use SHREDDAR directly, put it in front of a project's model traffic, or install its five-tool MCP boundary and portable Agent Skill in OpenClaw, Hermes, Codex, or Claude.

CHOOSE YOUR PATH ↓AGENT INSTALLS ↓API REFERENCE →
INDIVIDUAL
BROWSER LOCAL
PROJECT
SELF-HOSTED GATEWAY
AI + AGENT
MODEL / TOOL GUARD
A private document is divided into bounded SHREDDAR routes before sparse protected context reaches separate AI destinations
BOUNDARY ACTIVESOURCE → SHREDDAR → DESTINATIONONLY POLICY-APPROVED CONTEXT MOVES
WORKFLOW CONTROL / 01ADD THE BOUNDARY WHERE CONTEXT LEAVES.

CHOOSE YOUR ENTRY POINT

One boundary.
Three ways in.

Start in the browser, place the gateway inside a project, or mediate an agent's model and tool calls. Each route states exactly what is live now.

NO INSTALL / BROWSER LOCAL

Inspect before you share.

Use the working browser lab to reduce a document to the facts an AI task actually needs. The source and identity mapping remain in the current tab.

STATUS
LIVE IN PUBLIC ALPHA
RUNS IN
THIS BROWSER TAB
01 / SOURCEPRIVATE SOURCE
02 / CONTROLBROWSER SHREDDARPOLICY · MINIMIZE · RECEIPT
03 / DESTINATIONREVIEWED OUTPUT
  1. 01

    Choose a safe source

    Start with a synthetic case or document you are authorized to process. PDF, DOCX, TXT, Markdown, CSV, and JSON are extracted in browser memory within the published limits.

  2. 02

    State the exact task

    A narrow purpose lets SHREDDAR remove unrelated contact, financial, identity, and credential data.

  3. 03

    Review the outbound view

    Compare the private source, protected representation, simulated answer, local STITCH result, and metadata-only receipt.

  4. 04

    Send only after review

    The public lab does not call an external model. Copy the protected output only when it matches your intended boundary.

BOUNDARY PLAN
SOURCE     synthetic or authorized text
TASK       one explicit purpose
POLICY     balanced | strict | maximum
REVIEW     inspect OUTBOUND before use
EVIDENCE   retain the metadata-only receipt
No account, API key, wallet, or upload is required for the browser proof.
AVAILABLE NOW
  • Local PDF + DOCX extraction
  • Three privacy modes
  • Explicit outbound review
  • Verified evidence export
BOUNDARY / NEXT

Scanned documents, OCR, archives, rich-layout preservation, and production-sensitive handling remain self-hosted or held work.

OPEN THE APP LAB ↗

AGENT INSTALLS / ONE PRIVACY BOUNDARY

Keep your agent.
Add SHREDDAR.

OpenClaw, Hermes, Codex, Claude, and any MCP host use the same five safe tools. Copy a setup, download its config, and install one portable Agent Skill.

OCSHREDDAR / AGENT RUNTIME

MCP + SKILL READY

OpenClaw

Save SHREDDAR as an OpenClaw-managed MCP server, probe it before use, and install the reviewed privacy skill in the active workspace.

OPENCLAW MCP GUIDE ↗
  1. 01

    Export the local gateway URL and API key in the Gateway host environment.

  2. 02

    Add the stdio server with only the five agent-safe tools enabled.

  3. 03

    Probe the server before an agent turn so setup failures are visible.

  4. 04

    Install the portable skill and start a fresh session to refresh the skill snapshot.

COPY / RUN LOCALLY
export SHREDDAR_API_URL=http://127.0.0.1:8787
export SHREDDAR_API_KEY='replace-with-local-gateway-key'

openclaw mcp add shreddar \
  --command /ABSOLUTE/PATH/TO/shreddar/backend/.venv/bin/shreddar-mcp \
  --include 'analyze,protect,shard,get_receipt,get_usage'
openclaw mcp doctor shreddar --probe

mkdir -p ./shreddar-privacy
curl -fsS https://shreddar.pro/integrations/shreddar-privacy/SKILL.md -o ./shreddar-privacy/SKILL.md
openclaw skills install ./shreddar-privacy --as shreddar-privacy
openclaw skills check
VERIFY BEFORE USERun openclaw mcp doctor shreddar --probe, then /skill shreddar-privacy in a fresh session.
SURFACESTATUSWHAT EXISTS
MCP STDIO SERVERREADYFive typed tools; no reconstruction or Vault credentials in model context.
OPENCLAWREADYManaged MCP config plus reviewed workspace skill.
HERMES AGENTREADYNative MCP command plus direct Agent Skill install.
OPENAI CODEXCONFIG READYShared MCP config plus repo-scoped Agent Skill.
CLAUDE CODECONFIG READYProject .mcp.json plus repo-scoped Agent Skill.
OPENAI-COMPATIBLEREADYNon-streaming gateway adapter and privacy receipt.
VIRTUALS ACPHELDBlueprint only; identity, pricing and settlement are not activated.

THE OPERATING PATTERN

Same controls.
Every surface.

Integration changes where the privacy boundary runs—not the responsibility to define purpose, review output, and validate the deployment.

01

Start with purpose

Describe the job before moving context. A narrower task gives the compiler a stronger basis for removing unrelated facts.

02

Keep control local

Run detection, policy, identity mappings, and reconstruction inside the browser or infrastructure you trust.

03

Inspect every boundary

Review the protected representation before a model or tool receives it, then inspect returned output before STITCH.

04

Retain evidence, not source

Use privacy-minimized receipts for operations and debugging without turning prompts or Vault mappings into telemetry.

ALPHA BOUNDARY

Integrate what exists.
Label what comes next.

LIVE

Browser-local product lab

Deterministic detection, task-aware transformation, local reconstruction, and metadata-only receipts run in the current tab.

LIVE / SELF-HOSTED

OpenAI-compatible gateway

Authenticated non-streaming chat, protect, reconstruct, policy, Vault, output guard, and receipt routes run inside deployer-controlled infrastructure.

SELF-HOSTED READY

MCP + portable Agent Skill

A tested stdio server exposes five bounded tools. Copy and download setups are provided for OpenClaw, Hermes Agent, OpenAI Codex, and Claude Code.

HELD

Virtuals ACP + x402 settlement

The ACP blueprint is visible, but network identity, signer policy, posted rate, escrow, payment, and recovery evidence are required before activation.

TRY THE BROWSER LAB ↗REVIEW SECURITY →