Backlog

Next up

In Progress

Done

jhansi features

Sandboxes, runs, and the execution record persist across restarts. On startup jhansi reconciles what it has on record against what is actually running, closes out runs interrupted by the restart, and cleans up its own orphaned containers — leaving anything it did not create alone.

Planned

Survive a restart

jhansi.io

Sandboxes get a time-to-live and are reclaimed automatically once idle, so an abandoned session no longer holds a container and a working directory indefinitely. An optional absolute ceiling caps total lifetime regardless of activity.Expiry is recorded as a distinct event from user-initiated deletion. Stored run output is never removed by cleanup.

Sandbox expiry and automatic cleanup

Expose the facts your monitoring already knows how to consume: data directory writable, free and total space, isolation engine reachable. Plus a startup check so a misconfigured data directory fails at boot with a clear message instead of at the first run.jhansi exposes signals; your existing monitoring decides the thresholds. Included in the open-source install.

Health and diagnostics endpoint

Write files into a sandbox, read results back out, and list what a run produced. Write accepts multiple files in a single call, so an agent can push a whole set of generated files at once.Turns jhansi from "run this command" into "run this work" — write a script, execute it, fetch the output.

Run files in and out of a sandbox

Connect any MCP client — Claude among them — directly to your jhansi install and let it create sandboxes, run code, and move files, without writing a line of HTTP.Runs inside the same single binary. No extra process, no separate service.

MCP server

Every run's stdout and stderr are written to disk under the data directory, byte-for-byte matching the hashes recorded in its execution record — so a stored run can be verified rather than trusted.Run records live outside the sandbox directory and survive the sandbox that produced them.

Completed

Durable run output

<code>Execution events gain the detail that makes them evidence rather than a timeline: the command that ran, the exit code, duration, output sizes, and a SHA-256 of stdout and stderr.</code><code>Output contents never enter the event log — only their hashes.</code>