projectspg

Next-Gen Zero-Trust AI Gateway

Build what's next
on the Private Cloud

The high-performance AI privacy and routing layer. Real-time zero-knowledge de-identification, 10 sovereign regulatory compliance packs, and instant multi-provider LLM orchestration with sub-millisecond overhead.

ProjectSPG Architecture: Zero-Trust De-Identification, Sub-Millisecond Latency, 10+ Sovereign Regulatory Packs, Tested Across 1.94 Billion Tokens

The ProjectSPG Platform

End-to-end data de-identification, zero-retention privacy routing, and continuous enterprise compliance.

Automatic Data Redaction

Automatically redact customer PII and confidential records before prompts reach AI models, preventing data leaks without slowing development.

Context-Preserving Masking

Disguise sensitive names and numbers while preserving formatting and grammar, letting models reason accurately without seeing real data.

Global Privacy Scanner

Detect and shield regional identifiers across North America, Europe, Asia, Africa, and Latin America with zero manual setup.

Secure Enterprise Vault

Keep real data isolated in your dedicated vault, secured by enterprise encryption keys that only your organization controls.

How ProjectSPG Works

The 4-step sovereign privacy pipeline — from prompt interception to real-time rehydration.

STEP 01 • PROMPT

Prompt Ingestion & Edge Interception

<1MS SUB-MILLISECOND P99

STEP 02 • DE-IDENTIFY

Sovereign De-Identification & Tokenization

BYOK KMS AES-256

STEP 03 • INFERENCE

Zero-Trust Upstream AI Model Inference

GROQ • MISTRAL • OPENAI

STEP 04 • REHYDRATE

Real-Time Response & Streaming Rehydration

STREAMING SSE DETOKENIZATION

TESTED ACROSS

9.34 MILLION PROMPTS

1.94 BILLION TOKENS

Learn more about ProjectSPG

ALL BLOG POSTS

The
Open
Model
AI Stack

the MIGHT stack

models

∞

inference

• SGLang • vLLM • TRT-LLM ♥ ProjectSPG

gateways & routers

LiteLLM OpenRouter AI Gateway

harness

opencode pi Cursor

tools

MCP Skills

INFERENCE & SECURITY

The Open Source AI Stack: Why ProjectSPG Ranks #1

A deep dive into the open model AI stack and a ranked comparison of the leading AI security proxies across latency, reversibility, and sovereign compliance...

projectspg

Empirical Benchmark: 9.33M Prompts Evaluated with 100.00% Fidelity

BENCHMARK

Empirical Benchmark: 9,334,805 Prompts Evaluated with 100.00% Fidelity & 86µs Latency

Exhaustive empirical validation across 1.94B tokens proving 100.00% zero-loss de-identification, 17,735 prompts/sec throughput, and 13.3ms HTTPS latency...

projectspg

Global Sovereign AI Privacy: 109 Jurisdictions Supported

COMPLIANCE

Global Sovereign AI Privacy: 109 Jurisdictions Supported by ProjectSPG

Protecting sensitive enterprise data across GDPR, India DPDP, Singapore PDPA, HIPAA, and 109 sovereign national jurisdictions without code changes...

projectspg

Enterprise Guardrails & Threat Models

ENTERPRISE

Enterprise Guardrails for Regulated Industries: Healthcare, FinTech & Legal

Protecting clinical PHI under HIPAA, financial assets under PCI-DSS, attorney-client privilege, and defense telemetry across frontier LLMs with zero code changes...

PRIVATE BETA • BY INVITATION ONLY

Exclusive Sovereign Access

To guarantee zero-latency throughput, dedicated KMS hardware envelope isolation, and sovereign compliance oversight across our 109 supported jurisdictions, ProjectSPG is currently accessible on an invitation-only basis.

Apply for Private Access

We review requests daily to maintain isolated compute capacity. Approved organizations receive an invitation token within 24 hours.

Work / Corporate Email *

Company / Organization Website *

Primary Compliance Mandate

Estimated Monthly Token Volume

Start building on ProjectSPG

From real-time prompt de-identification to enterprise-grade privacy compliance

Playground

SYSTEM

USER

ProjectSPG Privacy Gateway Active

All 10 Packs Active

RESPONSE

0 tokens • 0.00s

Welcome to the Playground

• You can start by typing a prompt in the "User Message" field
• Click "Submit" (Or press Cmd + Enter) to get a response.
• When you're ready, click the " Add to Conversation" button to add the result to the messages.
• Use the "View Code" button to copy the code snippet to your project.

Don't forget to check out the documentation .

Target: Cloudflare Edge SSL Encrypted

PARAMETERS

Temperature

Max Completion Tokens

Reasoning

Stream

JSON Mode

Built-in tools

Browser Search

Code Interpreter

Moderation: safeguard

Top P

Seed

Stop Sequence

Template

API Keys

Manage your project API keys. Remember to keep your API keys safe to prevent unauthorized access.

Swipe table horizontally to view full API keys data

NAME	SECRET KEY	TIER	CREATED	LAST USED	EXPIRES	USAGE (24HRS)

Usage

View usage data for your project

Note: Data can be delayed by up to 15 minutes. All data shown in UTC time.

Total Spend $0.0000 USD

Projected cost calculation as if you were enrolled in billing. You will not be billed until you upgrade.

Model Activity Breakdown

Aggregate usage, token consumption, and privacy interception counts per supported model

Live Metering

Swipe table horizontally to inspect all metrics

Supported Model	Provider	Requests	Input Tokens	Output Tokens	Total Tokens	Protected Entities	Est. Spend

Logs

Show Errors Only

Swipe table horizontally to inspect all 11 telemetry columns

REQUEST TIME	MODEL	API KEY	CODE	TTFT	LATENCY	INPUT TOKENS	OUTPUT TOKENS	AUDIO SECONDS	REQUEST ID	ERROR

Page Size

Zero-Code Drop-In Privacy Layer

ProjectSPG API Reference

Integrate privacy-preserving, zero-data-leakage LLM intelligence into your stack. ProjectSPG functions as a wire-compatible, transparent drop-in replacement for the OpenAI SDK, LangChain, LlamaIndex, and native HTTP clients.

Gateway Base URLs

Primary API Gateway

https://api.projectspg.info/v1

Web Edge Gateway

https://projectspg.info/v1

Standard 1-Line Client Migration Compatible with OpenAI SDK v1.0+

Simply replace your client base URL with ProjectSPG. Every prompt is intercepted at sub-millisecond speeds, de-identified using cryptographic tokenization, sent to the model safely, and restored on return.

# Python from openai import OpenAI client = OpenAI( api_key="spg_live_your_key_here", base_url="https://api.projectspg.info/v1" # Drop-in ProjectSPG Gateway ) response = client.chat.completions.create( model="openai/gpt-oss-120b", messages=[{"role": "user", "content": "Schedule appointment for David at SSN: 123-45-6789"}] ) print(response.choices[0].message.content)

Authentication & Enterprise Privacy Headers

Authenticate all API requests via standard Bearer tokens and configure zero-knowledge encryption parameters.

HEADER	TYPE	REQUIRED	DESCRIPTION
Authorization	string	Required	Bearer token with your project API key: `Bearer spg_live_...`
Content-Type	string	Required	Must be `application/json` for JSON request bodies.
x-tokenization-mode	string	Optional	Privacy mode: `mask` (default: e.g. `[PERSON_1]`), `synthetic` (realistic fake data), or `fpe` (Format-Preserving Encryption).
x-detection-categories	string	Optional	Comma-separated regions or categories to scan (e.g. `north_america,european_union,financial,corporate` or `all`). Default: `all`.
x-vault-encryption-key	string	Optional	Client-provided 32-character AES-256-GCM encryption key. Zero plaintext is stored in the vault without this key.
x-custom-entities	string	Optional	JSON array or comma-separated list of proprietary keywords or project codenames to redact (e.g. `["ProjectTitan", "SecretAlgorithm"]`).

POST

/v1/chat/completions

Creates a privacy-sanitized chat completion. Ingests raw user prompts, masks 109-country PII/financial/health entities with cryptographic tokens at sub-millisecond edge speed, executes upstream inference, and rehydrates the model output back to the original entities. Supports real-time Server-Sent Events (SSE) streaming.

https://api.projectspg.info/v1/chat/completions

Request Body Parameters

PARAMETER	TYPE	DEFAULT	DESCRIPTION
model	string	-	Required. Model ID, e.g. `openai/gpt-oss-120b`, `codestral-2508`, `ministral-8b-2512`, `gemini-3.5-flash-lite`.
messages	array<object>	-	Required. Array of message objects, each containing `role` (`system`, `user`, or `assistant`) and `content` (string).
temperature	number	1.0	Sampling temperature between 0 and 2. Lower values result in more deterministic completions.
max_tokens	integer	2048	The maximum number of tokens to generate in the chat completion.
stream	boolean	false	If true, partial message deltas are streamed as Server-Sent Events (SSE) using the StreamTokenBuffer.
response_format	object	null	Set to `{ "type": "json_object" }` to enable JSON mode output.
seed	integer	null	Deterministic sampling seed for reproducible outputs.
stop	string \| array	null	Up to 4 sequences where the API will stop generating further tokens.

curl https://api.projectspg.info/v1/chat/completions \ -H "Authorization: Bearer spg_live_your_key_here" \ -H "Content-Type: application/json" \ -H "x-tokenization-mode: mask" \ -H "x-detection-categories: all" \ -d '{ "model": "openai/gpt-oss-120b", "messages": [ { "role": "system", "content": "You are a helpful customer support assistant." }, { "role": "user", "content": "Hello, my email is alice@company.com and my credit card is 4111-2222-3333-4444. Can you refund me?" } ], "temperature": 0.7, "max_tokens": 1024 }'

Example Response (200 OK)

{ "id": "chatcmpl-9x1u...abc", "object": "chat.completion", "created": 1727982000, "model": "openai/gpt-oss-120b", "choices": [ { "index": 0, "message": { "role": "assistant", "content": "I have initiated the refund for alice@company.com to your card ending in 4444." }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": 42, "completion_tokens": 28, "total_tokens": 70 }, "privacy": { "protected_prompt": "Hello, my email is [EMAIL_1] and my credit card is [CARD_1]. Can you refund me?", "entities_count": 2, "token_map": { "[EMAIL_1]": "alice@company.com", "[CARD_1]": "4111-2222-3333-4444" }, "engine_latency_us": 142 } }

POST

/v1/embeddings

Vector embedding privacy sanitization. Intercepts raw text or text batches, neutralizes PII and identifiers, and requests normalized vector embeddings from the upstream model without leaking raw data to vector database indexing pipelines.

https://api.projectspg.info/v1/embeddings

Parameters

PARAMETER	TYPE	REQUIRED	DESCRIPTION
input	string \| array<string>	Required	The text or array of strings to generate vector embeddings for.
model	string	Required	Embedding model identifier (e.g. `text-embedding-3-small`, `text-embedding-3-large`).

curl https://api.projectspg.info/v1/embeddings \ -H "Authorization: Bearer spg_live_your_key_here" \ -H "Content-Type: application/json" \ -d '{ "model": "text-embedding-3-small", "input": "Patient John Doe was admitted to hospital on 2026-08-12." }'

GET

/v1/models

Lists the catalog of available LLMs, including model IDs, upstream providers (Groq Cloud, Google AI Studio, Mistral AI), rate per 1M tokens, context windows, and operational tier tags.

https://api.projectspg.info/v1/models

curl https://api.projectspg.info/v1/models \ -H "Authorization: Bearer spg_live_your_key_here"

Example Response (200 OK)

{ "object": "list", "data": [ { "id": "openai/gpt-oss-120b", "object": "model", "owned_by": "groq", "provider": "Groq Cloud", "rate_per_1m_tokens": 0.15, "context_window": 131072 }, { "id": "codestral-2508", "object": "model", "owned_by": "mistral", "provider": "Mistral AI", "rate_per_1m_tokens": 0.30, "context_window": 262144 }, { "id": "ministral-8b-2512", "object": "model", "owned_by": "mistral", "provider": "Mistral AI", "rate_per_1m_tokens": 0.10, "context_window": 131072 } ] }

GET

/v1/models/{model}

Retrieves a single model instance and its operational configuration, verification status, and provider endpoint.

curl https://api.projectspg.info/v1/models/codestral-2508 \ -H "Authorization: Bearer spg_live_your_key_here"

POST

/api/tokenize

Standalone data de-identification & reversible tokenization. Neutralizes sensitive PII, PHI, financial cards, and custom keywords without invoking an external LLM. Use this to sanitize database rows, log pipelines, or document indexing before storage.

https://api.projectspg.info/api/tokenize

Parameters

PARAMETER	TYPE	DEFAULT	DESCRIPTION
text	string	-	Required. Raw input text to sanitize.
categories	array<string>	["all"]	Jurisdiction filters: `["north_america", "european_union", "financial"]`.
mode	string	"mask"	Tokenization scheme: `"mask"`, `"synthetic"`, or `"fpe"`.
sessionId	string	auto	Optional session ID to persist token mappings in the edge vault for later rehydration.
ttlSeconds	integer	86400	Time to live in seconds for the session mapping (default: 24 hours).

curl https://api.projectspg.info/api/tokenize \ -H "Authorization: Bearer spg_live_your_key_here" \ -H "Content-Type: application/json" \ -d '{ "text": "Transfer $5,000 from account 9876543210 to David Lee (SSN: 000-11-2222).", "mode": "mask", "categories": ["all"] }'

Example Response (200 OK)

{ "sanitizedText": "Transfer $5,000 from account [BANK_ACCOUNT_1] to [PERSON_1] (SSN: [SSN_1]).", "sessionId": "sess_9x2u1m7a", "entitiesCount": 3, "entitiesDetected": ["BANK_ACCOUNT", "PERSON", "SSN"], "tokenMap": { "[BANK_ACCOUNT_1]": "9876543210", "[PERSON_1]": "David Lee", "[SSN_1]": "000-11-2222" }, "engineLatencyUs": 84 }

POST

/api/detokenize

Reversible rehydration. Takes a sanitized text containing tokens (e.g. [PERSON_1]) and restores the original plaintext values using either the stored sessionId or an explicitly passed tokenMap.

https://api.projectspg.info/api/detokenize

curl https://api.projectspg.info/api/detokenize \ -H "Authorization: Bearer spg_live_your_key_here" \ -H "Content-Type: application/json" \ -d '{ "text": "Payment sent to [PERSON_1] at [BANK_ACCOUNT_1].", "sessionId": "sess_9x2u1m7a" }'

Example Response (200 OK)

{ "rehydratedText": "Payment sent to David Lee at 9876543210.", "entitiesRestored": 2, "sessionId": "sess_9x2u1m7a" }

GET

/v1/logs

Retrieves the recent API call telemetry logs strictly belonging to the authenticated API key or user account. Used for real-time metering, token usage tracking, and audit verification.

Query Parameters

PARAMETER	TYPE	DEFAULT	DESCRIPTION
limit	integer	50	Number of recent log records to return (maximum: 100).

curl "https://api.projectspg.info/v1/logs?limit=10" \ -H "Authorization: Bearer spg_live_your_key_here"

GET

/v1/audit/events

Retrieves structured, zero-plaintext audit events formatted for enterprise SIEM ingestion (Splunk, Datadog, AWS CloudWatch). Every event includes rule IDs, SHA-256 fingerprints, execution latencies, and KMS encryption flags.

curl "https://api.projectspg.info/v1/audit/events?limit=25" \ -H "Authorization: Bearer spg_live_your_key_here"

GET POST DELETE

API Key Management (/api/keys)

Programmatically provision, inspect, and revoke API keys for your infrastructure, continuous deployment pipelines, and microservices.

GET /api/keys

List all active keys for account

curl https://api.projectspg.info/api/keys \ -H "Authorization: Bearer <Firebase_ID_Token>"

POST /api/keys

Provision a new API key

curl -X POST https://api.projectspg.info/api/keys \ -H "Authorization: Bearer <Firebase_ID_Token>" \ -H "Content-Type: application/json" \ -d '{ "name": "Production Microservice", "tier": "free", "monthlyQuota": 10000 }'

DELETE /api/keys/{id}

Revoke an active key instantly

curl -X DELETE https://api.projectspg.info/api/keys/key_123456 \ -H "Authorization: Bearer <Firebase_ID_Token>"

GET

/health

Global edge health check and latency ping. Returns cluster operational status, response timestamp, Cloudflare Colo edge region, and active engine version.

curl https://api.projectspg.info/health

Example Response (200 OK)

{ "status": "healthy", "gateway": "ProjectSPG Privacy Gateway", "version": "1.2.0", "region": "HKG", "timestamp": "2026-10-04T02:45:00.000Z" }

Invitation Code Manager

Admin Console

Create, allocate, and monitor secure invitation codes for partners and enterprise clients to unlock Full Access.

Active Codes

Ready for redemption

Total Redemptions

Full Access upgrades granted

Admin Account

boruahpriyanuj2004@gmail.com

Root Developer Access

Generate New Invitation Code

Issued Invitation Codes

0 codes total

Code	Recipient / Notes	Redemptions	Status	Created Date	Actions
Loading invitation codes...

Build what's next on the Private Cloud

The ProjectSPG Platform

Automatic Data Redaction

Context-Preserving Masking

Global Privacy Scanner

Secure Enterprise Vault

One-Click AI Gateway

Multi-Model Freedom (BYOK)

Seamless Live Rehydration

Real-Time Security Audit Logs

Turnkey Regulatory Compliance

How ProjectSPG Works

Prompt Ingestion & Edge Interception

Sovereign De-Identification & Tokenization

Zero-Trust Upstream AI Model Inference

Real-Time Response & Streaming Rehydration

Learn more about ProjectSPG

The Open Model AI Stack

The Open Source AI Stack: Why ProjectSPG Ranks #1

Empirical Benchmark: 9,334,805 Prompts Evaluated with 100.00% Fidelity & 86µs Latency

Global Sovereign AI Privacy: 109 Jurisdictions Supported by ProjectSPG

Enterprise Guardrails for Regulated Industries: Healthcare, FinTech & Legal

Exclusive Sovereign Access

Apply for Private Access

Invitation Request Received

Redeem Access Token

Start building on ProjectSPG

Playground

Welcome to the Playground

API Keys

Metrics

Usage

Model Activity Breakdown

Logs

Batch

ProjectSPG API Reference

Gateway Base URLs

Authentication & Enterprise Privacy Headers

/v1/chat/completions

Request Body Parameters

Example Response (200 OK)

/v1/embeddings

Parameters

/v1/models

Example Response (200 OK)

/v1/models/{model}

/api/tokenize

Parameters

Example Response (200 OK)

/api/detokenize

Example Response (200 OK)

/v1/logs

Query Parameters

/v1/audit/events

API Key Management (/api/keys)

/health

Example Response (200 OK)

Invitation Code Manager

Generate New Invitation Code

Issued Invitation Codes

Gateway Parameters

Provider API Keys

Sign in to your account

Welcome to ProjectSPG

Private Invitation Preview

Build what's next
on the Private Cloud

The
Open
Model
AI Stack