Xilos
OPEN SOURCE AI SECURITY & PRIVACY

The securest way to AI — now open source.

Detect PII, block jailbreaks, and guard every AI call — before data ever reaches a provider. Xilos is the open source layer that sits between your applications and the LLMs. Self-host. Audit the code. Your infrastructure, your data, your rules.

Solo (single user) · Org (team, RBAC) · Xilos Cloud (managed)

install.sh
git clone https://github.com/Mill-Pond-Research/xilos-oss.git
cd xilos-oss
./install.sh
# Pick Solo or Org mode
# Open http://localhost:3001

Open Source

BSL 1.1 Licensed

Docker

One-command setup

PII Protection

Real-time detection

17ms

Cache Response Time

One line change. Everything inherits.

Point your existing agents at Xilos. One base_url change gets routing, caching, guardrails, and full cost tracking.

# Point Claude Code at Xilos
export ANTHROPIC_BASE_URL="https://api.xilos.ai/api/v1"
export ANTHROPIC_API_KEY=sk-xilos-...

# Claude Code now routes through Xilos:
# → Smart routing to best model
# → Semantic cache (50% lower costs)
# → PII detection & guardrails
# → MCP tools with audit trails

Run your AI through Xilos and it changes everything.

Protect your data with real-time PII detection and local guardrails. Enhance every query with prompt improvement and contextual knowledge. Save money by caching questions and answers. All open source, all on your infrastructure.

Multi-Model Routing

Route every query to the best model based on cost, latency, or capability. Define rules in natural language — no code required.

Semantic Cache

Cache responses for semantically similar queries. Cache hits bypass LLMs entirely — lower cost, faster responses, smarter over time.

Agent Guardrails

Local PII detection, jailbreak prevention, and content filters. No data leaves your infrastructure.

Workflow Orchestration

Chain LLMs sequentially or in parallel. Build multi-stage pipelines — search with one model, reason with another, summarize with a third.

Cost Visibility

Per-model, per-user, per-org cost tracking. Know exactly what your agents are spending, in real time.

Context Engine

Every answered query becomes a searchable knowledge asset. Your agents' collective experience compounds over time.

MCP Tool Integration

Connect any Model Context Protocol server. Web search, file access, code execution — all routed through Xilos with audit trails.

Audit & Compliance

Complete immutable trail of every AI interaction. Per-request logs with prompt, response, model, tokens, and cost.

Self-Host or Cloud

Run on your own infrastructure with Docker, or use Xilos Cloud for zero-ops managed hosting.

Self-host Solo or Org. Scale to Cloud.

Start with Solo for personal use — no login, no auth wall. Upgrade to Org for teams with RBAC and shared governance. Or let us host it with Cloud.

Solo

For personal use

Free
  • Single user, no login required
  • Self-host with Docker
  • All routing features
  • Semantic caching
  • Basic guardrails
Self-Host
Org

For teams

Free
  • Multi-user with RBAC
  • Email/password or SSO
  • Teams and departments
  • Shared governance
  • Invite flows and audit
Self-Host
Cloud

We host it. Zero-ops.

From $20/mo
  • Fully managed by MPR
  • Unlimited users
  • Enterprise SSO (SAML)
  • 99.9% SLA
  • Priority support
Try Xilos Cloud
OPEN SOURCE SECURITY

Audit the code. Trust nothing else.

Every AI provider wants you to trust their security promises. Xilos is different — the code is open. PII detection, jailbreak prevention, content filtering — read every line. Self-host on your infrastructure. Zero telemetry. Zero call-home. Free for Solo and Org use.

View on GitHub

BSL 1.1

License

Solo + Org

Free to use

No Telemetry

No call-home

How Xilos compares

The only open-source AI gateway that gives you total control.

FeatureXilosOpenRouterPortkeyHeliconeLiteLLMCloudflarePalo AltoMicrosoftOpenAIAnthropic
Open source
Self-hosted
Multi-model routing
Semantic cache
Local guardrails
Workflow engine
MCP tool integration
Cost tracking
Audit trails
Context Engine

✓ = native feature · — = not available · Xilos is the only open-source gateway combining all of the above

Don't want to run it yourself?

Xilos Cloud — the same gateway, hosted and managed by Mill Pond Research. Zero-ops, auto-scaling, enterprise controls.

  • Zero-ops, auto-scaling, updates handled
  • Teams, orgs, and enterprise controls
  • SOC 2, SLAs, priority support

Frequently asked questions

How is Xilos different from Portkey?
Xilos is fully open source (BSL), self-hostable, and includes a workflow engine, MCP tool integration, semantic cache, and context engine — features Portkey doesn't offer. Portkey is a hosted-only service with per-request pricing. Xilos gives you total control.
Can I self-host Xilos?
Yes. Run ./install.sh from the repo, pick Solo or Org mode, and you're up in minutes. Everything runs in Docker on your infrastructure. No telemetry, no call-home. Full docs at docs.xilos.ai.
What LLM providers do you support?
Xilos supports 18+ LLM providers including OpenAI, Anthropic, Google, DeepSeek, Groq, Mistral, Cohere, and any OpenAI-compatible endpoint. Add your own API keys and route between them freely.
What about data privacy?
Self-hosted: your data never leaves your infrastructure. Cloud: data is encrypted at rest and in transit. SOC 2 compliant. No training on your data.
Can I migrate from self-host to Cloud?
Yes. Export your configuration, routing rules, and context engine data. Import into Cloud. We'll help with the migration. No downtime required.
What is the BSL license?
The Business Source License (BSL 1.1) allows free use for Solo and Org deployments. It restricts use as a competing service. After the change date (typically 3 years), it becomes Apache 2.0. Self-hosted use is always free.

Secure your AI. No provider gets your data.

Clone the repo, run the installer, and your AI has a security layer. Free. Open source. Audit the code yourself.