Jnvoy
An open-source HTTP proxy tool that intercepts and detects risks such as PII, prompt injection, and MCP tool poisoning before LLM calls. It performs reversible redaction on requests and then forwards them to upstream models, so security hardening can be applied without modifying business code.
Timeline (last 90 days)
Jnvoy is an open-source HTTP proxy deployed between applications and LLM services, performing security checks such as PII detection, masking, and injection detection before data leaves the local network.
When using Jnvoy, developers only need to replace the request URL and change the original Anthropic SDK call to an httpx.post request, adding a provider field to the request body to specify the upstream model provider, with no need to change business logic.
Jnvoy identifies PII entities on the request path and replaces them with reversible tokens, and on the response path restores the placeholders to their original values based on a local mapping table, transparently to the caller.
The security precondition of reversible token replacement is that the mapping table exists only inside the proxy, and in multi-tenant scenarios each API key corresponds to an independent mapping space to avoid data crossover between tenants.
Jnvoy's PII detection is based on Microsoft Presidio, supports more than 20 entity types, and additionally adds UK-localized recognizers covering NHS numbers, NI numbers, UK postcodes, and sort codes.
Jnvoy's prompt injection detection targets three categories of attack patterns: jailbreaks, instruction overrides, and system prompt extraction.
Jnvoy's MCP security module additionally detects tool poisoning and secret leakage in tool responses.
Jnvoy writes audit logs to PostgreSQL and supports one-click generation of PDFs of FCA and EU AI Act compliance reports.
Jnvoy's semantic cache is said to save 20% to 40% of LLM call costs.
The Jnvoy project contains 164 automated tests, provides a one-click Docker startup stack, and includes a Slack connector for protecting AI bot messages.
3 more timeline events
All Facts (13)
Jnvoy is an open-source HTTP proxy deployed between applications and LLM services, performing security checks such as PII detection, masking, and injection detection before data leaves the local network.
50%UnverifiedWhen using Jnvoy, developers only need to replace the request URL and change the original Anthropic SDK call to an httpx.post request, adding a provider field to the request body to specify the upstream model provider, with no need to change business logic.
50%UnverifiedJnvoy identifies PII entities on the request path and replaces them with reversible tokens, and on the response path restores the placeholders to their original values based on a local mapping table, transparently to the caller.
50%UnverifiedThe security precondition of reversible token replacement is that the mapping table exists only inside the proxy, and in multi-tenant scenarios each API key corresponds to an independent mapping space to avoid data crossover between tenants.
50%UnverifiedJnvoy's PII detection is based on Microsoft Presidio, supports more than 20 entity types, and additionally adds UK-localized recognizers covering NHS numbers, NI numbers, UK postcodes, and sort codes.
50%UnverifiedJnvoy's prompt injection detection targets three categories of attack patterns: jailbreaks, instruction overrides, and system prompt extraction.
50%UnverifiedJnvoy's MCP security module additionally detects tool poisoning and secret leakage in tool responses.
50%UnverifiedJnvoy writes audit logs to PostgreSQL and supports one-click generation of PDFs of FCA and EU AI Act compliance reports.
50%UnverifiedJnvoy's semantic cache is said to save 20% to 40% of LLM call costs.
50%UnverifiedThe Jnvoy project contains 164 automated tests, provides a one-click Docker startup stack, and includes a Slack connector for protecting AI bot messages.
50%UnverifiedAll requests passing through a local proxy inevitably add latency, PII detection requires computing resources, and throughput performance under high concurrency is the key variable for actual deployment.
50%UnverifiedIf the proxy itself is compromised, the reversible token mapping table's restoration capability becomes an attack surface, so the proxy's level of security protection determines overall security.
50%UnverifiedGDPR and UK FCA compliance requirements impose clear constraints on personal data processing, but most teams do not proactively build a separate detection layer during rapid iteration, which is Jnvoy's market entry point.
50%