islo

Deterministic harness-optimization simulation on five toy tasks. Scripted agent and pattern proposer; Islo backend stub.

10-agent GTM AI team on OpenClaw, sharing a Cognee+FalkorDB knowledge graph, deployed to an Islo long-running sandbox, with a Mission Control HTTP UI

Project page for a deterministic harness-optimization simulation: scope, source, and reproducible toy results.

Demo: Bazel 9 build of a pinned pydantic/monty snapshot, with a measured remote-cache run (121.1s cold, 15.7s cached) and an Islo sandbox build.

Minimal Great Expectations (GX Core 1.x) data-quality demo with one check that fails by design; runs locally or in islo.dev and crabbox sandboxes.

POC field guide (June 2026): when native caches, BuildKit, or Incredibuild Build Cache speed up GitHub Actions on fresh or warm runners.

Talk slides, Apache Airflow Monthly Virtual Town Hall (July 10, 2026): one ephemeral microVM sandbox per Airflow task try.

Demo: scan all 24 GRCh38 chromosomes by snapshotting one warm islo.dev sandbox and forking it per chromosome, checked against a known CpG-island gradient.

Demo: GA-style prompt evolution over parallel islo.dev sandboxes with a Pokémon Crystal-style viewer; the published results are a scripted playback.

Experiment: an oracle loop asks a 1B local model (gemma3:1b) to write x86_64 NASM for Wolfram's Rule 30 in an islo VM; 3 iterations, none correct yet.

Demo page (May 2026): scoring Repo2RLEnv pr_diff tasks on an islo.dev sandbox via openclaw/crabbox. The upstream PR was closed unmerged.

Unofficial interactive demo of the Inferoa agent harness: a client-side token-cost simulation plus receipts from real vLLM runs in islo.dev sandboxes.

Landing page for unity-loop, a May 2026 proof of concept in which Claude proposes and judges CSS variants of a fixed Unity WebGL scene on islo.dev sandboxes.

Project page for a June 2026 demo: the Odysseus AI workspace (odysseus-dev) on throwaway islo.dev microVMs via openclaw/crabbox.

Demo: Python script that renders a 10-second vertical video of a scripted four-agent call, with ElevenLabs or local text-to-speech voices.

Synthetic reproduction of the Avellaneda-Stoikov vs symmetric market-maker comparison from a Stanford MSE448 report, run in an Islo sandbox.

Demo World Cup prediction market (no real money): three Node.js microservices and a gateway with LMSR pricing, built with Claude Code to run on islo.dev.

May 2026 demo: one Wordle env from adithya-s-k/RL_Envs_101 in four RL frameworks, two served from short-lived islo.dev sandboxes; transcripts are scripted.

Alpha third-party Apache Airflow provider that runs tasks in ephemeral sandboxes (local, Daytona, E2B, Modal, islo) via SandboxOperator or SandboxExecutor.

Toy LongMemEval-style demo: three scripted memory strategies on three handcrafted questions, run in islo.dev sandboxes. Not an official LongMemEval score.