Source linked

TensorZero Goes Dark: 11.5k-Star LLMOps Platform Archived Overnight

github.com@systems_wire4 hours ago·Artificial Intelligence·4 comments

TensorZero, an open-source LLMOps platform handling ~1% of global LLM API spend, was archived on June 12, 2026 - leaving a gap in gateway, observability, and optimization tooling.

tensorzerollmopsopen sourcellm gatewayobservabilityai infrastructure

TensorZero, the open-source LLMOps platform that claimed to fuel ~1% of global LLM API spend, was archived on June 12, 2026 with zero warning — the repo is now read-only. 11.5k stars, 838 forks, 4,100 commits, and a $7.3M seed round (per external reports) didn't save it from the owner hitting the archive button.

What TensorZero Actually Did

TensorZero wasn't another LLM wrapper. It combined a Rust-based gateway, observability, evaluation, and optimization into one platform. The gateway delivered <1ms p99 latency overhead at 10k+ QPS — fast enough for production traffic. It supported every major provider: Anthropic, AWS Bedrock, AWS SageMaker, Azure, DeepSeek, Fireworks, GCP Vertex AI, Google AI Studio, Groq, Hyperbolic, Mistral, OpenAI, OpenRouter, SGLang, TGI, Together AI, vLLM, xAI (Grok), plus any OpenAI-compatible API like Ollama. Integrations required one Docker container and a base_url swap.

Observability wasn't an afterthought. TensorZero stored every inference and feedback in your own database, exposed programmatically or through its UI. It promised AI-assisted debugging and data labeling as upcoming features — now frozen by the archive. The optimization engine used supervised fine-tuning, RLHF, GEPA (automated prompt engineering), and dynamic in-context learning. A/B testing, routing, fallbacks, and retries were built in.

The Autopilot and the Sudden Stop

TensorZero Autopilot, described as an automated AI engineer, analyzed observability data, set up evals, and ran A/B tests to improve LLM agent performance. That feature was highlighted as new in the README, likely the last major push before the project went silent.

No public explanation accompanied the archive. The repository simply states “archived by the owner on Jun 12, 2026.” For teams that depended on TensorZero for gateway, monitoring, and optimization, this is an immediate migration trigger. The platform was used by companies from startups to Fortune 10, and its loss leaves a hole in the open-source LLMOps stack that competitors like Langfuse, Helicone, or Portkey will have to fill — or the community will have to fork.

Source: AI OSS tool repo goes archived over night after raising $7.3M Seed
Domain: github.com

Read original source ->

External source stays available while the OJO article and comment thread stay local.

More in Artificial Intelligence

view topic
SFT, Not RL, Determines Gemini's Safety Properties

Google DeepMind found that supervised fine-tuning alone reproduces nearly all safety behavior of production Gemini models, making RL a secondary lever for alignment.

Selling KV Caches Cuts AI Prefill Costs 50x

Loading a precomputed KV cache matches prefill token-for-token at 9-50x lower compute, but shipping the cache over the network costs more than it saves; provider-side hosting unlocks the real margin.

NVIDIA Blackwell Runs 20x More Agentic AI Agents per Megawatt Than Hopper

NVIDIA's GB300 NVL72 platform tops the first agentic AI infrastructure benchmark, delivering 20x more concurrent agents per megawatt than Hopper-because agentic workloads are a fundamentally different beast from chat.

21 Zero-Days in FFmpeg: $1K Agent Outperforms $10K Mythos Scan

depthfirst's autonomous security agent discovered 21 zero-day vulnerabilities in FFmpeg, including bugs latent for 23 years, at a fraction of the cost of Anthropic's Mythos analysis.

Comments load interactively on the live page.