Stockyard vs TensorZero

Two self-hosted LLM gateways with different philosophies. TensorZero optimizes automatically. Stockyard gives you manual control with more built-in features.

TensorZero is a Rust-based LLM gateway backed by FirstMark and Bessemer ($7.3M seed). It is fully free and open source under Apache 2.0. The standout feature is TensorZero Autopilot, which automatically optimizes your LLM setup based on observed performance.

Stockyard is a Go-based proxy with 40+ providers, 76 middleware modules, and embedded SQLite. It focuses on operational features like caching, rate limiting, guardrails, and audit trails that TensorZero does not include.

Both are self-hosted and both are fast. The question is whether you want automatic optimization or manual control with more built-in tooling.

FeatureStockyardTensorZero
LanguageGo (compiled binary)Rust (compiled binary)
DeploymentSingle binary, embedded SQLiteDocker, Postgres + ClickHouse optional
External depsNoneNone required (Postgres/ClickHouse optional for persistence)
Open sourceProxy: Apache 2.0 / Platform: BSL 1.1Apache 2.0 (11.2k stars)
PricingFree unlimited, paid from $0.99/mo per toolFully free, no paid tier
Providers40+~15
Latency overhead~400ns (76 modules)<1ms p99
Caching✓ Built-inNone
Rate limiting✓ Built-inNone
Failover✓ Built-inFallback support
Guardrails✓ 76 middleware modulesNone
Auto-optimizationManual (Drover routing)✓ Autopilot (automatic)
A/B testing✓ Tack Room✓ Experimentation
EvaluationsVia Tack Room✓ Built-in evaluations
Audit trail✓ Hash-chainedNone
Red-team testing✓ FeralNone
VC backingBootstrapped$7.3M seed (FirstMark, Bessemer)

Data reflects publicly available documentation as of March 2026.

Why Stockyard

More built-in features. TensorZero is a gateway and optimization engine. Stockyard is a gateway plus caching, rate limiting, guardrails, PII redaction, audit trail, red-team testing, prompt management, and request replay. If you need any of these, you would need to add separate tools on top of TensorZero.

More providers. Stockyard supports 40+ providers out of the box. TensorZero supports approximately 15. If you use providers like Cohere, xAI, or smaller regional providers, Stockyard covers them.

Zero external dependencies. TensorZero works without Postgres and ClickHouse, but then you lose observability data persistence. Stockyard persists everything to embedded SQLite automatically.

When TensorZero might be better

If automatic optimization is your priority, TensorZero is purpose-built for it. Autopilot observes your request patterns and automatically optimizes your LLM configuration. Stockyard has Drover for cost-optimized routing, but it is not as deeply integrated as TensorZero's optimization loop.

TensorZero is fully free and fully Apache 2.0. No BSL, no paid tiers, no enterprise upsell. If budget is zero and you want a gateway with an optimization loop, TensorZero costs nothing.

TensorZero has Fortune 10 customers and $7.3M in VC backing. It is a well-funded project with strong institutional support. Stockyard is bootstrapped by a solo founder.

The bottom line

TensorZero is the better choice if auto-optimization is your primary need and you do not need caching, rate limiting, or guardrails. Stockyard is the better choice if you want a full operational toolkit in a single binary with zero dependencies.

Try Stockyard → Proxy-only mode See pricing
Also compare: vs Helicone  ·  vs LiteLLM  ·  vs Portkey  ·  vs OpenRouter  ·  vs Langfuse
Explore: Self-hosted proxy · Best self-hosted proxy · OpenAI proxy · Install guide
Stockyard also makes 150 focused self-hosted tools — browse the catalog or get everything for $29/mo.