Vowdo Agent 0.2.0rc5 — Native TypeScript/Deno Preview

Vowdo Agent 0.2.0rc5 — native TypeScript/Deno preview

A native, local workflow for work you can inspect and results you can verify.
Download the Linux x64 archive; both executable binaries are included.
No Python, Deno installation, account or API key is required for the CSV workflow.

Supported: compare CSV files, inspect saved runs, export and verify content;
scripted fixture execution with recovery and budget accounting. Improvement
commands demonstrate a fixture protocol/lifecycle, not measured model gains.

Live model execution, validation of applied runner configuration, TUI,
arbitrary untrusted execution and other platforms are outside this preview.
The controller has host-wide filesystem/process authority; the worker
revokes ambient grants and checks effective denial. Worker memory has no
enforced OS cap. See the source quickstart and security documentation.

Verify downloaded files with SHA256SUMS before extracting. MANIFEST.json
binds binaries to the public source commit and build toolchain. Tests on
a fresh public-source export: 60 passed / 0 failed / 1 optional runner skip;
with the real Choose runner configured: 61 passed / 0 failed. The compiled
no-Python journey and source/history secret scans passed.

Apache-2.0 core; embedded runtime license texts are included. The earlier
Python v0.1.0rc1 release and its assets remain available unchanged.

Packaging correction: the SQLite attribution wrapper now names the Vowdo
executables. Its legal dedication, both binary hashes, the source archive
and the release tag are unchanged.

Vowdo Agent 0.2.0rc4 — Historical Native Preview

Historical build, preserved with its original artifact filenames and checksums.

Vowdo Agent 0.2.0rc4 — native TypeScript/Deno preview

A native, local workflow for work you can inspect and results you can verify.
Download the Linux x64 archive; both executable binaries are included.
No Python, Deno installation, account or API key is required for the CSV workflow.

Supported: compare CSV files, inspect saved runs, export and verify content;
scripted fixture execution with recovery and budget accounting. Improvement
commands demonstrate a fixture protocol/lifecycle, not measured model gains.

Live model execution, validation of applied runner configuration, TUI,
arbitrary untrusted execution and other platforms are outside this preview.
The controller has host-wide filesystem/process authority; the worker
revokes ambient grants and checks effective denial. Worker memory has no
enforced OS cap. See the source quickstart and security documentation.

Verify downloaded files with SHA256SUMS before extracting. MANIFEST.json
binds binaries to the public source commit and build toolchain. Tests on
a fresh public-source export: 55 passed / 0 failed / 1 optional runner skip;
with the real Choose runner configured: 56 passed / 0 failed. The compiled
no-Python journey and source/history secret scans passed.

Apache-2.0 core; embedded runtime license texts are included. The earlier
Python v0.1.0rc1 release and its assets remain available unchanged.

Vowdo Agent 0.1.0rc1 — Historical Python Reference

Historical build, preserved with its original artifact filenames and checksums.

Vowdo Agent 0.1.0rc1 — developer preview

A local-first starting point for work you can inspect and results you can
verify. This Apache-2.0 preview supports task-only workspace initialization,
CSV reconciliation with an explicit key, saved-run inspection, and native
export with byte-level verification.

Supported

• Changed values, missing rows, duplicate-key ambiguity and schema differences.
• Immutable input snapshots and saved task state.
• Export manifests, fresh-process verification and tamper rejection.
• Synthetic examples, including spaces and Chinese filenames.

Experimental and deferred

The TUI, JAZ demonstrations, controlled-improvement adapters and loopback
provider simulations are experimental. External integration prerequisites
apply. Live AI execution, autonomous business operations, cloud accounts,
payments and unverified platforms are deferred. This release makes no claim
of measured model-quality improvements or a production sandbox.

Verification

• Linux/Python 3.13 standalone suite: 821 passed, 36 explicitly skipped
  external-runner tests; no failures.
• Three loopback simulation tests passed with the configured external runner.
• Ruff and mypy passed; source/history and artifact secret scans passed.
• Fresh installed-wheel task journey and tampered-export rejection passed.

See INSTALL.md for setup and SHA256SUMS for downloadable-file integrity.
The repository includes contribution guidance and public architecture/protocol
documentation. No PyPI publication is claimed.