← oncell.aiUSE CASE

AI Coding Agents

Build AI coding agents where each user gets their own sandboxed environment — their own filesystem, shell, git repo, and search index. No Docker, no Kubernetes, no infrastructure code.

The problem

AI coding agents need per-user isolation. Each user has their own codebase, their own git history, their own test results. You can't share a single container. But provisioning a container per user means managing Docker, orchestrating lifecycle, syncing state to S3, and handling crashes.

How OnCell solves it

One API call creates an isolated environment for each user. Your agent code runs inside with direct access to the filesystem, shell, database, and search — all on local NVMe at 0ms latency.

const cell = await oncell.cells.create({
  customerId: "user-123",
  agent: codingAgentCode,
});

// Inside the agent:
ctx.shell("git clone https://github.com/user/repo /cell/work");
const files = ctx.search.query("authentication logic");
ctx.shell("npm test");
ctx.stream({ type: "test_result", passed: true });

What your agent gets

Filesystem — read, write, delete files. Clone repos, install dependencies, run builds. Everything persists across sessions.

Shell — run any command. Git, npm, pip, make. Full shell access with output capture.

Search— semantic code search across the user's codebase. Find relevant files for the LLM context window.

Database — track conversation history, agent state, task progress. Per-user, local, fast.

Streaming — stream progress to the frontend in real-time. Show the user what the agent is doing as it works.

Crash recovery — if the agent crashes mid-task, it resumes from the last successful step. No lost work.

Example architectures

Cursor-style coding assistant — user connects a repo, agent reads files, makes edits, runs tests. Each user has their own clone.

Bolt/Lovable-style app generator — agent generates code, previews it at a live URL, iterates on user feedback. Each project is an isolated cell.

Code review agent — agent checks out PRs, runs analysis, posts comments. Per-repo isolation with persistent state.

Start building

One API call. Per-user filesystem, shell, database, search.