HomeBlogPricingCareersDocsGitHubSlack community
TENSORLAKE·ENGINEERING LOG / FULL ARCHIVE

The full archive.

Every post we’ve published — engineering notes, benchmarks, model comparisons, and announcements. The curated feed lives on the main log.

POSTS74
Archive —CURRENT LOG
55 POSTS · NEWEST FIRST
Jul 22, 2026Engineering

Run Devin Outposts on Tensorlake Sandboxes.Cognition keeps the agent loop; execution runs inside a Tensorlake sandbox you control. We built the integration, and you can clone and run it today.

Shanshan WangApplied AI · Tensorlake7 min
Jul 20, 2026Community builds

Crabbox Adds Tensorlake as a Delegated-Run ProviderOpenClaw's Crabbox — warm a box, sync the diff, run the suite — drives Firecracker microVMs through the tensorlake CLI, so one flag puts your test run in a sandbox.

OpenClawCommunity · crabbox.sh2 min
Jul 13, 2026Engineering

Firecracker disk snapshots in O(changed bytes), not O(disk size)We added copy-on-write image layering inside Firecracker's block device, and the disk got faster, not slower. Snapshots pause the VM for tens of milliseconds and cost only what changed, so a running Postgres barely notices them, at any disk size.

Diptanu Gon ChoudhuryCEO / Co-founder10 min
Jul 11, 2026Community builds

Building Stateful AI Agents That Survive Session KillsAgent amnesia, solved: 84ms cold starts, snapshots as memory that carries prior experience, and forked sandboxes to evaluate candidate solutions in parallel.

Gowtham BoyinaCommunity · Towards AI10 min
Jul 10, 2026Community builds

Your Agent Benchmark Is Lying to YouAn 87% reported pass rate fell to 33% once verified — 53% of passes were fraudulent. Fork-per-task microVM isolation plus hidden verifiers make evals trustworthy.

Sebastian BuzduganCommunity · Data Science Collective12 min
Jul 10, 2026Community builds

Why Your AI Experiments Keep Starting From Scratch (And How Tensorlake Fixes It)Warm up the environment once, memory-snapshot it, then fork N workers that resume mid-execution — instead of paying the same 40s setup tax on every run.

Divy YadavCommunity · Towards AI12 min
Jul 9, 2026Community builds

I Ran 5 AI Agents in Parallel on Tensorlake. The Isolation Held. Here Is How I Built It.Five agents in parallel, one deliberately crashed mid-run — the others finished clean with no defensive code, just per-agent microVMs. Registered images cut boot from 19.49s to 4.51s.

Darshan DagaCommunity · Towards AI12 min
Jul 9, 2026Engineering

Zero-copy TLS ingress with kTLS and splice(2) for sandboxes

Diptanu Gon ChoudhuryCEO & Co-founder8 min
Jul 6, 2026Community builds

Building Long-Running Claude-Managed Agents: Why State Matters More Than ComputeAn orchestrator/worker architecture on sandboxes that suspend when idle — checkpoint state and stop paying for compute instead of keeping boxes alive.

Divy YadavCommunity · Towards AI11 min
Jun 24, 2026Engineering

Accelerating Harbor with Tensorlake.Tensorlake runs the real task Docker images for Harbor on microVM sandboxes. On Terminal-Bench 2.1 the oracle passes all 89 tasks in 36m 44s — every image published, cold starts in seconds.

Shanshan WangApplied AI · Tensorlake6 min
Jun 17, 2026Community builds

Demo: A Sandbox Agent Built on TensorlakeAn AI coding assistant that writes, runs, and exports code inside a Tensorlake sandbox — demo on X, code on GitHub.

Sundaram Kumar JhaCommunity · X1 min
Jun 17, 2026Community builds

Snapshot Once, Rollout a Thousand Times: A Practical RL Setup for Coding AgentsSnapshot the environment once and fork every rollout from it: 8,000 RL rollouts drop from ~20 hours of naive environment rebuilds to ~4.4 hours.

Sebastian BuzduganCommunity · Data Science Collective11 min
Jun 16, 2026Community builds

Tensorlake Is Now a Sandbox Provider in DeputiesDeputies, an open-source control plane for background agents, added Tensorlake as a sandbox provider — after it met every criterion from the project's own "what makes a good sandbox" post.

Sid PalasCommunity · X1 min
Jun 16, 2026Community builds

Why Most Multi-Agent AI Systems Waste 90% of Their Time (And How to Fix It)The bottleneck isn't concurrency — it's setup overhead. At 50 agents that's 4,500s spent preparing to work vs 50s with memory snapshots.

Divy YadavCommunity · Towards AI15 min
Jun 16, 2026Community builds

Building a Stateful Code Interpreter with Tensorlake SandboxesA build-along tutorial: from a toy code interpreter to one that remembers across sessions, suspends, snapshots, and forks — all on the free tier.

Rohan MistryCommunity · Towards AI9 min
Jun 11, 2026Community builds

Your AI Agent Is One rm -rf Away From Disaster. Here Is What I Found After 5 Sandbox Experiments.Five isolation experiments running hostile, LLM-generated code. All four isolation boundaries held, and boots averaged 1.121s with only 148ms of variance.

Darshan DagaCommunity · Towards AI8 min
Jun 5, 2026Engineering

Tensorlake Sandboxes as the runtime for Claude Managed Agents.Run Claude Managed Agents on Tensorlake Sandboxes: you keep the control of self-hosting, and drop the always-on box. The short version — the server goes away.

Shanshan WangApplied AI · Tensorlake7 min
Jun 4, 2026Community builds

8 AI Agent Sandboxes, Ranked — and Why the Fastest One Will Wreck Your Agent in ProductionEight sandboxes benchmarked head-to-head. Tensorlake places fourth on raw cold-start speed but first on consistency: a 0.09s median-to-P99 spread vs 0.37s+ elsewhere.

Chew Loong NianCommunity · Towards AI11 min
May 26, 2026Community builds

I Built a Stateful Research Agent Inside a Sandbox. Here's What the Numbers Actually Looked Like.A research agent that keeps its state in a Tensorlake sandbox: it survives restarts and resumes mid-task, with real latency numbers throughout.

Divy YadavCommunity · Towards AI9 min
May 18, 2026Research

Claude Opus 4.7 vs. Kimi K2.6: Real-World Coding TestA practical comparison between the premium closed frontier model and the cheaper open-weight agentic coder, on the same two-part coding task.

12 min
May 13, 2026Engineering

Filter and rank: how we schedule sandboxes across every cloudOne primitive — filter, then rank — runs through statefulness, multi-cloud, BYOC, and the data path. A tour of the scheduler behind Tensorlake's sandbox fleet.

Diptanu Gon ChoudhuryCEO & Co-founder14 min
May 1, 2026Engineering

Browser Harness: direct Chrome DevTools Protocol access and a self-healing harness for browser agentsBrowser-use strips their harness to 592 lines and gives the LLM a raw WebSocket to Chrome's DevTools Protocol. When the agent hits a gap, it writes a helper function — and saves it for future runs.

Antonio Jimeno YepesEngineering3 min
Apr 30, 2026Engineering

How scaffold design affects coding agent benchmark scores: lessons from DroidContextual instruction injection, per-model tool schemas, and planning/execution splits — the harness decisions behind Droid's 77.3% on Terminal-Bench.

Antonio Jimeno YepesEngineering3 min
Apr 29, 2026Engineering

Hermes: the coding agent that gets better the more you use itMost coding agents are stateless. Hermes uses a closed learning loop where the agent creates skills from experience, improves them during use, and builds a persistent model of who you are across sessions.

Antonio Jimeno YepesEngineering3 min
Apr 29, 2026Engineering

Pi: a coding agent with efficient system promptingPi keeps its entire system prompt — including all tool definitions — under 1,000 tokens, a 10× reduction over tools like Claude Code or Cline.

Antonio Jimeno YepesEngineering2 min
Apr 28, 2026Engineering

ForgeCode: top open source coding agent in Terminal-Bench@2.0ForgeCode reaches 81.8% on Terminal-Bench 2.0 with both Claude Opus 4.6 and GPT-5.4 — a look at what the harness is doing that the model isn't.

Antonio Jimeno YepesEngineering3 min
Apr 27, 2026Engineering

Building Sandboxes for Computer UseComputer Use is a tiny loop. The hard part is building the boring, reproducible desktop around it.

BERNARD KOLOBARASoftware Engineer · Tensorlake8 min
Apr 24, 2026Engineering

Starting hundreds of sandboxes in parallel, and the design that makes it possible.We replaced reconciliation loops with durable command outbox to scale our sandbox scheduler to start 1000s of sandboxes every second

David CalaveraSoftware Engineer · Tensorlake 9 min
Apr 22, 2026Engineering

Tensorlake is now an official Harbor environment runtimeHarbor defines and evaluates terminal tasks. Tensorlake provides the MicroVM execution layer. Together they are a full evaluation stack for CLI agents.

Tensorlake TeamEngineering7 min
Apr 16, 2026Concepts

Suspend vs. snapshot: pause a sandbox, or save it for reuse?One is a pause button, the other is a save file. Same state, different question — and the answer shapes your cost model, your fan-out pattern, and which failures you can recover from.

Diptanu Gon ChoudhuryCEO / Co-founder12 min
Apr 14, 2026Engineering

How we got to 5,000,000 sandboxes per projectA walk through the scheduler rewrite, the snapshot format, and the two syscalls that freed us from the noisy-neighbor tax.

Diptanu Gon ChoudhuryCEO / Co-founder12 min
Apr 09, 2026Engineering

Sub-second cold starts for stateful microVMsSnapshot-restore beats boot every time. How we cut P50 wake-up to 80ms.

Priya RaoStaff Engineer9 min
Apr 3, 2026Engineering

Autoresearch on steroids with sandboxesAn LLM agent can propose incremental training-script improvements, but safely executing untrusted code requires isolated sandboxes with resource limits — that's where Tensorlake comes in. Here's the overnight hill-climb, end to end.

Tensorlake TeamEngineering9 min
Apr 02, 2026Product

RL environments are generally availableShip your reward model alongside the same sandbox that runs inference — no separate training harness.

Mark LiuHead of Product4 min
Mar 27, 2026Engineering

Durable agent loops without a queueA checkpoint per step, at 2.4ms. Why we stopped writing to Redis and started writing to disk.

Sasha VolkovaRuntime Engineer11 min
Mar 25, 2026Engineering

Harbor x TensorLake: Infrastructure for Agentic EvalsTensorLake is now integrated as a first-class environment provider in Harbor, enabling scalable agent evaluation with secure ephemeral MicroVMs.

Tensorlake TeamEngineering2 min
Mar 21, 2026Research

Filesystem benchmarks: sandboxed vs. bare metalSeven workloads, four runtimes. Most sandboxes leave 40% of your disk throughput on the table.

Noah KimResearch18 min
Mar 08, 2026Product

BYOC: run Tensorlake in your own AWS accountSingle-tenant control plane, PrivateLink, and data residency — without giving up the managed experience.

Mark LiuHead of Product6 min
Feb 28, 2026Company

We raised our Series A32M to build the sandbox layer every agent needs. A note on why, whats next, and who were hiring.

Diptanu Gon ChoudhuryCEO / Co-founder3 min
Feb 19, 2026Engineering

Five patterns for tool execution in production agentsEphemeral vs. persistent, per-tenant vs. shared, and when to pay the cold-start tax on purpose.

Sasha VolkovaRuntime Engineer8 min
Feb 11, 2026Research

Paged KV caches across sandbox boundariesA small experiment in sharing attention caches between sibling sandboxes without breaking isolation.

Noah KimResearch22 min
Feb 9, 2026Research

Claude Opus 4.6 vs GPT 5.3 CodexGPT 5.3 Codex excels at fast execution and agentic coding, while Claude Opus 4.6 stands out in reasoning-heavy and long-context tasks.

Tensorlake TeamEngineering6 min
Feb 04, 2026Company

SOC 2 Type II, and a primer on our security modelWhat we attest to, what we dont, and the exact kernel boundaries youre trusting.

Alex ParkSecurity5 min
Feb 2, 2026Research

OpenCode: The Best Claude Code AlternativeMost developers now use AI agents for coding tasks, but Claude Code's managed approach and constraints around model selection, pricing, and customization have spurred interest in open-source alternatives like OpenCode, which offers greater flexibility and control.

Tensorlake TeamEngineering9 min
Jan 29, 2026Engineering

The End of Database-Backed Workflow Engines: Building GraphRAG on Object StorageGraphRAG ingestion needs fan-out parallelism and durable retries that database-backed workflow engines can't give it; object storage can.

Tensorlake TeamEngineering8 min
Jan 28, 2026Product

Python 3.12 is now the default imagePlus a new base image with uv pre-cached for 6x faster dependency installs.

Mark LiuHead of Product2 min
Jan 19, 2026Engineering

Building HackerNews Podcast Generator with Gemini 3, ElevenlabsThis article shows how to build a simple podcast generator that turns Hacker News posts into short audio summaries using a single Tensorlake Application.

Tensorlake TeamEngineering8 min
Jan 18, 2026Research

OpenAI GPT-5.2-Codex (high) vs. Claude Opus 4.5 vs. Gemini 3 Pro: Real-World Coding comparisonThree leading AI models were tested on real coding tasks in an 8K+ star repository to compare code quality, reliability, and practical performance.

Tensorlake TeamEngineering10 min
Jan 14, 2026Research

Claude Cowork: Architecture, Capabilities, and Usage OverviewClaude Cowork is an agentic task execution mode allowing Claude to plan and complete multi-step work on your behalf.

Tensorlake TeamEngineering8 min
Jan 8, 2026Engineering

Building an Outage Agent That Handles Repeated Production IncidentsAn outage agent built on Tensorlake Applications triages alerts, investigates with Groq and Exa, and only escalates when a human is actually needed.

Tensorlake TeamEngineering9 min
Dec 28, 2025Research

The Next Enterprise Platform Isn't Data-Driven, It's Context-DrivenContext graphs capture the reasoning behind decisions, not just their outputs, making AI-driven workflows auditable without replacing existing systems of record.

Tensorlake TeamEngineering11 min
Dec 15, 2025Research

Gemini 3 vs GPT-5.2: Detailed Coding ComparisonGemini 3 Pro is fast and multimodal but produces simpler output; GPT-5.2 trades speed for more reliable, production-ready code.

Tensorlake TeamEngineering8 min
Dec 13, 2025Research

I Tried Letting Antigravity Build An Agent For Me. Here's What Actually HappenedWe let Antigravity build a guest checkout flow and abandoned-cart recovery emails end to end, and tracked where it held up and where it didn't.

Tensorlake TeamEngineering8 min
Dec 9, 2025Research

OpenAI GPT-5.2 Codex vs. Gemini 3 Pro vs Opus 4.5: Coding comparisonGPT-5.2 Codex, Gemini 3 Pro, and Claude Opus 4.5 build a Minecraft clone, a Figma dashboard, and solve a hard LeetCode problem, with very different results.

Tensorlake TeamEngineering8 min
Dec 9, 2025Engineering

TOON vs JSON: A Token-Optimized Data Format for Reducing LLM CostsTOON strips the repetitive syntax out of JSON for LLM prompts, cutting token usage while improving structured extraction accuracy.

Tensorlake TeamEngineering8 min
Legacy —DOCUMENT AI ERA
19 POSTS

Posts from before Tensorlake became the sandbox cloud for AI agents, kept online for reference. For what we build today, start with the current log.

Jan 20, 2026Engineering

Agentic Table MergingAgentic Table Merging reconstructs tables split across pages and columns by reasoning over content, not just geometry.

Tensorlake TeamEngineering8 min
Jan 9, 2026Engineering

Agentic Chart ExtractionAgentic Chart Extraction turns chart images into structured JSON series data, ready for analytics and BI pipelines.

Tensorlake TeamEngineering6 min
Dec 5, 2025Research

Gemini 3 OCR - Quick FindingsGemini 3 has capable OCR but delivers unstructured output; Tensorlake adds precise page slicing and well-organized JSON without manual cleanup.

Tensorlake TeamEngineering2 min
Dec 4, 2025Engineering

How Tensorlake Solved the DOCX Tracked Changes Problem for Legal TechTensorlake DocumentAI merges Word's tracked-changes XML with PDF-level spatial layout, preserving full audit trails for legal tech.

Tensorlake TeamEngineering5 min
Dec 2, 2025Product

Gemini 3 is Now Available as an OCR Model in TensorlakeGemini 3 is now available as an OCR engine inside Tensorlake's Document Ingestion API.

Tensorlake TeamEngineering3 min
Nov 5, 2025Research

Benchmarking the Most Reliable Document Parsing APITensorlake's document parsing model hits 91.7% accuracy on real enterprise documents, ahead of Azure, AWS Textract, and open-source alternatives.

Tensorlake TeamEngineering8 min
Sep 28, 2025Engineering

Precise Data Extraction: Pattern-Based Partitioning for Structured ExtractionPattern-based partitioning extracts from specific document sections instead of the whole page, so layout changes don't break your pipeline.

Tensorlake TeamEngineering3 min
Sep 25, 2025Engineering

Building Clean, Schema-Enforced Pipelines with Tensorlake + OutlinesCombining Tensorlake's structured parsing with Outlines' schema-enforced generation eliminates malformed JSON before it ever reaches your pipeline.

Tensorlake TeamEngineering5 min
Sep 18, 2025Engineering

Citation-Aware RAG: How to add Fine Grained Citations in Retrieval and Response SynthesisCitations are table stakes for agentic applications — without them, users can't trust the output.

Tensorlake TeamEngineering8 min
Sep 9, 2025Engineering

Parse and Retrieve Dense Tables Accurately with TensorlakeTensorlake preserves structure in dense, multi-page tables, returning DataFrames with bounding boxes so retrieval stays accurate and explainable.

Tensorlake TeamEngineering6 min
Sep 3, 2025Product

Verify Structured Output with Field-Level CitationsEvery field extracted by Tensorlake's Structured Extraction can now be traced back to its exact page and bounding box.

Tensorlake TeamEngineering6 min
Aug 22, 2025Engineering

Fix Broken Context in RAG with Tensorlake + ChonkieTensorlake's hierarchy-aware parsing plus Chonkie's chunking keeps RAG context coherent instead of chopping documents into meaningless pieces.

Tensorlake TeamEngineering6 min
Aug 19, 2025Engineering

Accelerate Advanced RAG with TensorlakeProduction RAG needs a retrieval plan over fresh, structured context, not just top-N cosine similarity over a stale corpus.

Tensorlake TeamEngineering8 min
Aug 11, 2025Engineering

AI Tagging for Page-Level Metadata with Tensorlake Page ClassificationPage Classification turns unstructured documents into page-level metadata that CRMs, RAG pipelines, and compliance workflows can act on directly.

Tensorlake TeamEngineering2 min
Jul 31, 2025Product

Page Classification: Smarter, Safer Structured ExtractionPage Classification labels every page up front, so extraction only runs where it's actually relevant.

Tensorlake TeamEngineering6 min
Jul 23, 2025Product

Unlocking Smarter RAG with Qdrant + Tensorlake: Structured Filters Meet Semantic SearchPairing Tensorlake's document understanding with Qdrant's hybrid search unlocks retrieval that filters on real document structure, not just vectors.

Tensorlake TeamEngineering8 min
Jun 11, 2025Product

LangChain + Tensorlake: Unlocking Document Understanding for AgentsTensorlake and LangChain partnered to bring reliable document parsing and understanding to LangGraph agents.

Tensorlake TeamEngineering7 min
May 28, 2025Product

Signature Detection in Tensorlake: Catch what's missing, trigger what's nextSignature Detection finds and contextualizes signatures in any document, turning a missing signature into an automated workflow trigger.

Tensorlake TeamEngineering6 min
May 15, 2025Product

Tensorlake Cloud: Ingest, Structure, Orchestrate Without Losing a ByteTensorlake Cloud is a fully managed platform for turning unstructured documents into structured, AI-ready data with zero dropped fields.

Tensorlake TeamEngineering7 min
◆ FIELD NOTES — WEEKLY

Engineering posts, in your inbox.

One dispatch per week from the Tensorlake team — runtime deep-dives, product updates, and the occasional benchmark that surprised us.