# Siphyric

*/Startups/Siphyric*

## Startup Overview

This security middleware intercepts outbound requests to large language models and tokenizes sensitive data payloads before they leave the corporate network. It replaces personally identifiable information and proprietary secrets with secure tokens prior to external routing. Upon receiving the model's response, the system detokenizes the output back into its original context for the application layer.

Engineering and security teams building AI applications face a strict data exposure barrier: sending regulated data to third-party endpoints violates compliance and governance policies. Relying on the data retention promises of API providers or attempting to maintain custom proxy filters drains resources and creates persistent vulnerability gaps. The software removes this risk by neutralizing the data locally before it ever reaches the public internet.

Standard interventions like Cloudflare AI Gateway, Nightfall AI, or brittle in-house proxy servers force tradeoffs involving centralized trust, network latency, or deep vendor lock-in. This architecture bypasses those limitations by remaining both strictly client-side encrypted and completely model-agnostic by design. Teams secure their prompts at the source and route sanitized requests to any commercial or open-source model without exposing raw data to intermediaries.

## Startup Founding Hypothesis

**Approach**: that tokenizes sensitive payloads before external LLM routing
**Competitors**:
- [Cloudflare AI Gateway](/Competitors/Cloudflare_AI_Gateway)
- [Nightfall AI](/Competitors/Nightfall_AI)
- [in-house proxy servers](/Competitors/in-house_proxy_servers)
**Differentiator2x2**: both client-side encrypted and completely model-agnostic by design

## Startup Solution Coordinate

**Solution**: [Siphyric Token Proxy](/Software/Siphyric_Token_Proxy)

## Startup Position2x2

```mermaid
quadrantChart
    title Siphyric Position
    x-axis Vendor-Locked --> Model-Agnostic
    y-axis Server-Side Gateway --> Client-Side Encrypted
    Cloudflare AI Gateway: [0.8, 0.2]
    Nightfall AI: [0.6, 0.3]
    in-house proxy servers: [0.7, 0.4]
    Siphyric: [0.9, 0.9]
```

## Startup Offer

**Proof**:
- Aim to secure daily LLM traffic for mid-market fintechs without altering model latency
- Target zero-leakage API routing for healthcare providers querying patient symptom data
- Design to pass rigorous infosec audits by operating as a strictly zero-knowledge broker
**Tiers**:
- Name: Developer Metered · Price: ~$0.02–$0.05 per 1,000 requests · Inclusions: Pay-as-you-go access to the tokenization proxy, standard PII redaction templates, and multi-model routing endpoints for individuals or small teams.
- Name: Growth Deployment · Price: ~$300–$600/mo · Inclusions: Up to 5 million requests per month, custom redaction rules, Bring Your Own Key (BYOK) architecture, and dedicated onboarding support.
- Name: Enterprise Edge · Price: ~$2,000–$5,000/mo · Inclusions: High-volume tiered pricing, dedicated isolated proxy instances, custom SLA guarantees, and enterprise SSO integrations.
**Guarantee**: If sensitive payloads pass through the proxy without matching the configured tokenization rules due to a system failure, Siphyric credits the full month's usage and provides a dedicated engineering review to patch the rule.
**Business Function**: ProvideService
**Objection Handlers**:
- Adds too much latency to real-time chat: Designed for <10ms overhead by utilizing lightweight, edge-optimized tokenization dictionaries.
- The LLM will lose reasoning context if data is masked: Uses semantic placeholders (e.g., substituting a real name with '[Person_A]') rather than hashes, preserving structural logic for the model.
- We can build a regex proxy in-house: Maintaining multi-model API specs, format preservation, and constantly evolving PII edge-cases internally consumes significant dedicated engineering time.
- Who controls the encryption keys?: Architecture is completely client-side encrypted; you hold the keys, ensuring Siphyric processes routing but never accesses plaintext.
**Pricing Architecture**: UsageMeter
**Agent Checkout Support**:
- agentic-commerce-protocol

## Startup Brand

**Voice**: Clinical and precise, anchored entirely in cryptographic certainty.
**Tagline**: Route prompts to any LLM without leaking sensitive data.
**Icon Concept**: envelope
**Palette Intent**: institutional-cool
**Visual Identity**: Deep navy blues and slate grays project cryptographic authority, paired with monospaced typography to reflect the developer-first tokenization environment.
**Archetype Reference**: the-sage

## Startup Buyer Chain

**Chain**: Siphyric → Enterprise DevSecOps → AI Application Developers → Enterprise End-Users
**Gtm Motion**: Acquires initial usage through bottom-up developer adoption of a lightweight local proxy for masking PII during prototyping, expanding into enterprise-wide contracts when security teams mandate centralized audit logs and custom KMS integrations for production LLM deployments.
**Agent Channel**: Designed to expose a secure routing tool via the Model Context Protocol (MCP) registry, intending to allow autonomous enterprise agents to dynamically discover and pipe their outgoing prompts through a compliant tokenization endpoint.
**Primary Channel**: Developer community discovery via GitHub open-source repositories and technical blog content targeting searches for 'LLM PII redaction proxy' and 'client-side encrypted OpenAI gateway'.

## Startup Customer Journey

```mermaid
flowchart LR
A[Developer GitHub Repo] --> B[Local PII Proxy]
B --> C[Redacted Tokenization Endpoint]
C --> D[Production Edge Instance]
D --> E[Custom KMS Architecture]
E --> F[SecOps Audit Log]
```

## Startup Proof Points

_Illustrative — target and order-of-magnitude estimate figures, not an achieved track record (this Thing is concept-stage)._

**Pilot Goals**:
- A 14-day staging environment integration routing 1 million requests to prove <10ms latency overhead and zero leakage of seeded PII data.
- A 30-day shadow deployment for an internal AI assistant to validate that the semantic placeholder substitution preserves prompt response quality compared to unredacted queries.
**Target Metrics**:
- Target: <10ms latency overhead added per proxied API request.
- Aim: 100% preservation of model reasoning context using semantic placeholders instead of hashes.
- Target: 0 instances of sensitive payload leakage during third-party penetration testing.
- Aim: >95% reduction in internal engineering hours previously dedicated to maintaining custom LLM regex parsers.
**Target Case Studies**:
- A mid-market fintech company routing daily LLM traffic through a zero-knowledge proxy to prevent financial data leakage without increasing model response times.
- A regional healthcare network substituting patient symptom data with semantic placeholders to query external models securely while preserving structural logic.
- A B2B SaaS engineering team replacing an internal regex-based redaction tool with a managed tokenization proxy to eliminate the engineering hours spent updating PII edge-cases.
**Testimonial Targets**:
- Chief Information Security Officer validating that the Bring Your Own Key (BYOK) architecture ensures the proxy routes traffic without ever accessing plaintext.
- Lead AI Engineer confirming that semantic placeholders maintain the LLM's reasoning context for complex multi-turn chat completions.
- VP of Engineering praising the immediate out-of-the-box compatibility with constantly evolving multi-model API specs.

## Startup Top Risks

**Risks**:
- Severity: existential · Description: Cloud infrastructure providers bundle native zero-trust LLM proxy gateways directly into their enterprise AI offerings, rendering a standalone routing layer obsolete. · Mitigation Status: unmitigated
- Severity: high · Description: Tokenization of critical entities strips semantic context from prompts, causing severe degradation in downstream LLM reasoning and response accuracy. · Mitigation Status: in-progress
- Severity: moderate · Description: Strict client-side deployment requirements introduce complex encryption key management hurdles that stall enterprise sales cycles. · Mitigation Status: in-progress
- Severity: moderate · Description: Established data security competitors replicate the client-side encryption architecture, eroding the primary technical differentiator. · Mitigation Status: unmitigated

## Startup Competitors

- [Cloudflare AI Gateway](/Competitors/Cloudflare_AI_Gateway) — Incumbent
- [Nightfall AI](/Competitors/Nightfall_AI) — DLP Provider
- [In-House Proxy Servers](/Competitors/In-House_Proxy_Servers) — Status Quo
- [Lakera Guard](/Competitors/Lakera_Guard) — AI Security Platform
- [Private AI](/Competitors/Private_AI) — Redaction Engine
- [Portkey AI Gateway](/Competitors/Portkey_AI_Gateway) — Routing Platform

## Startup Solution Stack

- [Payload Tokenization Service](/Services/Payload_Tokenization_Service) — Service-as-Software
- [Model Routing Agent](/Agents/Model_Routing_Agent) — Agent
- [Client Encryption SDK](/Software/Client_Encryption_SDK) — Software
- [Agnostic Proxy API](/Software/Agnostic_Proxy_API) — Software

## Startup Story Brand

**Hero**:
- **Need**: to be the technical architect who enables innovation while maintaining zero-leakage security standards
- **Want**: to utilize external LLMs without leaking sensitive customer PII or PHI
- **Identity**: the engineering lead at a mid-market fintech or healthcare provider
**Plan**:
- Step: Select templates · Detail: Choose from standard PII redaction or HIPAA-compliant rules to define what data gets masked.
- Step: Validate logic · Detail: Test how placeholders like [Person_A] maintain model reasoning without revealing plaintext identities.
- Step: Deploy proxy · Detail: Route your traffic through our zero-knowledge broker to secure every prompt automatically.
**Guide**:
- **Empathy**: Does your LLM integration still risk exposing sensitive payloads during real-time inference?
**Problem**:
- **Villain**: unprotected API routing
- **External**: Sending raw patient data or financial records to OpenAI or Anthropic violates HIPAA and SOC2 compliance immediately
- **Internal**: You feel like you are gambling with the company's reputation every time a developer hits an endpoint
- **Philosophical**: Enterprise data was built for internal utility, not for training third-party models.
**Success**: Ship LLM-powered features in weeks instead of months, backed by a cryptographically secure routing layer that never sees your plaintext data.
**One Liner**: Instead of building risky in-house regex proxies, Siphyric tokenizes sensitive payloads at the edge — enabling secure multi-model routing without leaking data.
**Positioning**:
- **So That**: route prompts to any LLM without exposing sensitive PII
- **Unlike**: manual in-house proxy servers
- **For Whom**: engineering leads at regulated fintechs and healthcare providers
- **Category**: Zero-knowledge AI Security Gateway
**Call To Action**:
- **Direct**: Provision a proxy
- **Transitional**: View tokenization schema
**Failure Stakes**:
- Compromised patient privacy
- Failed infosec audits
- Costly regulatory fines
**Transformation**:
- **To**: one of the few engineering leads who deploy secure AI at scale
- **From**: a cautious gatekeeper blocking AI adoption
**Controlling Idea**: Data security and AI innovation must coexist through cryptographic tokenization.

## Startup Token Hero

**Genre**: founding-hypothesis
**Rendered**: Instead of building risky in-house regex proxies, Siphyric tokenizes sensitive payloads at the edge — enabling secure multi-model routing without leaking data.
**Mechanism**: spine-derived-v1
**Template Id**: spine-founding-hypothesis
**Vocab Fingerprint**: 5283be238f7f40fc

## Startup Token Positioning

**Genre**: moore-positioning
**Rendered**: Zero-knowledge AI Security Gateway for engineering leads at regulated fintechs and healthcare providers. Unlike manual in-house proxy servers — route prompts to any LLM without exposing sensitive PII.
**Mechanism**: spine-derived-v1
**Template Id**: spine-moore-positioning
**Vocab Fingerprint**: 6c5eb9041aca8cc3

## Startup Token Pitch Deck

**Genre**: pitch-deck
**Rendered**: Problem: Sending raw patient data or financial records to OpenAI or Anthropic violates HIPAA and SOC2 compliance immediately
Solution: Instead of building risky in-house regex proxies, Siphyric tokenizes sensitive payloads at the edge — enabling secure multi-model routing without leaking data.
Customer: engineering leads at regulated fintechs and healthcare providers
Unlike: manual in-house proxy servers
**Mechanism**: spine-derived-v1
**Template Id**: spine-pitch-deck
**Vocab Fingerprint**: 4f59830470fbb4ff

## Startup Token M E D D P I C C

**Pain**: Sending raw patient data or financial records to OpenAI or Anthropic violates HIPAA and SOC2 compliance immediately
**Metrics**: Target: Ship LLM-powered features in weeks instead of months, backed by a cryptographically secure routing layer that never sees your plaintext data.
**Rendered**: Pain: Sending raw patient data or financial records to OpenAI or Anthropic violates HIPAA and SOC2 compliance immediately
Economic buyer: Enterprise DevSecOps
Metrics: Target: Ship LLM-powered features in weeks instead of months, backed by a cryptographically secure routing layer that never sees your plaintext data.
Competition: manual in-house proxy servers
**Mechanism**: spine-derived-v1
**Competition**: manual in-house proxy servers
**Economic Buyer**: Enterprise DevSecOps
**Vocab Fingerprint**: 5dc91e6451329c09

## Startup Token Cold Email

**Genre**: cold-email
**Rendered**: Subject: Zero-knowledge AI Security Gateway for engineering leads at regulated fintechs and healthcare providers

engineering leads at regulated fintechs and healthcare providers — Sending raw patient data or financial records to OpenAI or Anthropic violates HIPAA and SOC2 compliance immediately Instead of building risky in-house regex proxies, Siphyric tokenizes sensitive payloads at the edge — enabling secure multi-model routing without leaking data.
**Mechanism**: spine-derived-v1
**Template Id**: spine-cold-email
**Vocab Fingerprint**: c92b6393ba498717

## Startup Token Agent Spec

**Genre**: ai-agent-spec
**Rendered**: Zero-knowledge AI Security Gateway. Instead of building risky in-house regex proxies, Siphyric tokenizes sensitive payloads at the edge — enabling secure multi-model routing without leaking data. Serves engineering leads at regulated fintechs and healthcare providers.
**Mechanism**: spine-derived-v1
**Template Id**: spine-ai-agent-spec
**Vocab Fingerprint**: 663bc43f509e1142

## Neighborhood

### Candidate solutions

- [Expansion Pipeline Bottlenecks](/Problems/Expansion_Pipeline_Bottlenecks) — candidate solution for · Problems

### What it offers

- [Siphyric Token Proxy](/Software/Siphyric_Token_Proxy) — offers · Software

### Composed of

- [Payload Tokenization Service](/Services/Payload_Tokenization_Service) — composes · Services
- [Model Routing Agent](/Agents/Model_Routing_Agent) — composes · Agents
- [Client Encryption SDK](/Software/Client_Encryption_SDK) — composes · Software
- [Agnostic Proxy API](/Software/Agnostic_Proxy_API) — composes · Software

### Embodies

- [Software](/Theses/Software) — embodies · Theses

### Competitors

- [Nightfall AI](/Competitors/Nightfall_AI) — competes with · Competitors
- [In-House Proxy Servers](/Competitors/In-House_Proxy_Servers) — competes with · Competitors
- [Lakera Guard](/Competitors/Lakera_Guard) — competes with · Competitors
- [Private AI](/Competitors/Private_AI) — competes with · Competitors
- [Portkey AI Gateway](/Competitors/Portkey_AI_Gateway) — competes with · Competitors
- [Cloudflare AI Gateway](/Competitors/Cloudflare_AI_Gateway) — competes with · Competitors

### Similar Startups

- [Ciphermuri](/Startups/Ciphermuri) — similar · Startups
- [Safetyrouter](/Startups/Safetyrouter) — similar · Startups
- [Guardoster](/Startups/Guardoster) — similar · Startups
- [Datalock](/Startups/Datalock) — similar · Startups
- [Validategate](/Startups/Validategate) — similar · Startups
- [Shorepoint](/Startups/Shorepoint) — similar · Startups
- [Sensept](/Startups/Sensept) — similar · Startups
- [Gleamharbor](/Startups/Gleamharbor) — similar · Startups
- [Walledpulse](/Startups/Walledpulse) — similar · Startups
- [Embergate](/Startups/Embergate) — similar · Startups
- [Logreg](/Startups/Logreg) — similar · Startups
- [Bridgeshade](/Startups/Bridgeshade) — similar · Startups
- [Anontext](/Startups/Anontext) — similar · Startups
- [Stagesplit](/api/md.md.md.md/Problems/Global_IP_Data_Residency/Startups/Stagesplit) — similar · Startups
- [Bedractable](/Startups/Bedractable) — similar · Startups
- [In-House Sanitization Scripts](/Startups/In-House_Sanitization_Scripts) — similar · Startups

### Similar Opportunities

- [Support Privacy Firewall](/Opportunities/Support_Privacy_Firewall) — similar · Opportunities
- [AI API Scrubbing for Cloud](/Opportunities/AI_API_Scrubbing_for_Cloud) — similar · Opportunities
- [Contextual PII Firewall](/Opportunities/Contextual_PII_Firewall) — similar · Opportunities
- [Document Sanitization Layer](/api/md.md/Opportunities/Document_Sanitization_Layer) — similar · Opportunities