Tako VM v0.1.4 — complete Python SDK, reliability layer, and end-to-end hardening

[las7]

We've shipped Tako VM v0.1.4 — the largest release since the project's first PyPI publish. It graduates the Python SDK to a complete, production-grade client and hardens the execution engine end-to-end for durability, traceability, and security.

pip install --upgrade tako-vm

Why this release matters

If you're running untrusted code from AI agents, the two things that bite you in production are reliability (did the job actually run exactly once?) and observability (when it fails, can you tell why?). v0.1.4 is squarely aimed at both.

A complete Python SDK

Full API parity with the server — synchronous typed execution, the entire async job lifecycle (submit, get_status, get_result, cancel, rerun, fork), execution history, and metadata — now reachable both from the TakoVM client and the flat tako_vm.* module-level API:

import tako_vm

tako_vm.configure("https://tako.internal", headers={"X-API-Key": KEY})
job_id = tako_vm.submit(my_func, MyInput(1, 2))
result = tako_vm.get_result(job_id, timeout=30)

A built-in reliability layer

Transport retries, auto-generated idempotency keys for retry-safe submission, an X-Correlation-ID on every request (exposed on results and exceptions), and a structured exception taxonomy. Ambiguous network failures no longer mean double-executed code or silent data loss.

Verbose-on-failure everywhere

Every failure path — server, workers, sandbox, and SDK — now surfaces a correlated, actionable error instead of a vague status.

Also in this release

• Durability: sync /execute runs off the event loop and persists a record; idempotent retries with a fresh container per attempt; watchdog that honors per-job-type budgets; startup reconciliation of stale jobs.
• Security: opt-in API-key auth, DLQ payload redaction with configurable TTL, symlink rejection in artifact collection, and a published security policy + threat model.
• Operability: CLI --workers/--reload/host-port fixes, direct execution of pre-built job-type images, labeled containers with reliable orphan cleanup.

Links

• 📦 PyPI: https://pypi.org/project/tako-vm/0.1.4/
• 📝 Release notes: https://github.com/las7/TakoVM/releases/tag/v0.1.4
• 📜 Full changelog: https://github.com/las7/TakoVM/blob/main/CHANGELOG.md
• 🔁 Diff since v0.1.3: v0.1.3...v0.1.4

Known limitation

The async worker pool is still per-process and in-memory; multi-worker deployments are warned about at startup and remain an architectural constraint we plan to revisit.

Questions, feedback, or upgrade issues? Reply here or open a discussion in Q&A. 🐙