Map / Sandboxing / Islo
Islo
islo.dev ↗ · LinkedIn ↗ · Open on the map
SandboxingAI IdentityObservability & Governance
Capabilities
Secure, long-running sandbox for AI coding agents: each agent gets its own dedicated, persistent cloud environment (a real Linux microVM, not docker-in-docker) that runs real services, survives reconnects, and resumes with the same filesystem, history, and context. Granular policy controls govern each agent access to codebases, infrastructure, and external systems such as GitHub and Jira; a built-in vault keeps secrets out of the model by injecting credentials only at egress with per-request audit, so live keys never reach the agent or sandbox memory. Adds agent identity management, guardrails, human approval checkpoints, observability, cost control, and reproducible environments; partners with the Harbor Framework for reproducible agent benchmarks and reinforcement-learning evaluations with reward-hacking defense. Spin up from Python, TypeScript, Go, or the CLI.
Sensors / integration
Cloud-hosted isolated execution environments (real Linux microVMs) provisioned per agent; deployable standalone or via the Incredibuild SDLC platform; SDKs for Python, TypeScript, and Go plus a CLI; an egress-layer secrets vault that rewrites and injects credentials at the network boundary; integrates with Incredibuild acceleration for compute-heavy build and test workflows.
Protections
Per-agent isolation that decouples AI execution from developer machines and contains blast radius; granular policy enforcement over what each agent can access (codebases, infrastructure, GitHub, Jira); secret isolation via egress-time credential injection so keys never touch agent or sandbox memory, with a continuous per-request audit chain; human approval checkpoints for autonomous runs; guardrails, observability, and cost controls managed centrally.
Differentiator
Built by Incredibuild, the Israeli SDLC-acceleration company serving 600+ customers including Microsoft, Take-Two, and Barclays, and launched in 2026 — bringing enterprise execution-control and build-acceleration heritage to agent sandboxing. Thesis that every AI agent needs its own computer: agents that otherwise run on a developer laptop die when the lid closes and inherit the developer full permissions. Distinguished from pure execution-infra sandboxes such as Daytona by its egress-time secrets-vault model and tight coupling to Incredibuild acceleration and SDLC governance; a Harbor Framework partnership targets RL and benchmark workloads with reward-hacking defense.
Team & funding
- Founders
- Uri Mishol, Uri Shaham
Other Sandboxing companies
ARMO · Autonomous · Blaxel · Canyon Road · Daytona · Duality Technologies · Edera · Galene.AI · GuardiAgent · I-THAM · Jozu · Keyboard · See all →
Last updated · Get the “Listed on” badge