Repository navigation
v0.3.2
Wombat v0.3.2
This maintenance release reduces derived-index storage and cached-view memory while preserving exact accounting and historical evidence.
Highlights
- Read usage summaries and task rankings from exact persisted aggregates. Historical measurements and operation details load on demand; the recent 30-day aggregate cache has a byte limit even during heavy usage.
- Keep historical accounting and supported evidence available after source logs disappear. Fixed views retain their own committed facts and prices, including late results and corrections.
- Reduce repeated parsing, encoding, and association work during rebuilding and appends. Cached restoration shares task and turn metadata instead of retaining another complete ledger.
- Bound concurrent query reductions and spill larger working sets to private temporary storage. Resource limits return explicit errors rather than silently truncated totals.
- A fixed synthetic fixture measured more than tenfold complete-index reduction against the original uncompressed layout. This is fixture-specific evidence, not a guarantee for every source corpus.
Install or update
macOS and Linux:
curl -fsSL https://raw.githubusercontent.com/YannByte/wombat/main/scripts/install/install.sh | shWindows PowerShell:
& ([scriptblock]::Create((irm https://raw.githubusercontent.com/YannByte/wombat/main/scripts/install/install.ps1)))Update an installer-managed copy:
wombat update --check
wombat updateSupported archives: darwin-arm64, darwin-x64, linux-x64, linux-arm64, win32-x64.
Known limitations
- Codex is the only production data source in this release; support for other agents is planned.
- API costs are estimates. Unpriced or partial data stays explicit and is separate from subscription allowance.
- The derived-index format changes in this release. Older indexes can return INDEX_UNSUPPORTED_VERSION and are preserved without migration or automatic deletion. Follow the documented format recovery and preserve user decisions. A different WOMBAT_DATA_HOME creates an isolated store; it does not repair the old store or transfer its decisions.
- Timing describes recorded activity and can be partial or provisional. It does not establish why work was slow, code quality, token savings, or model performance.
- A sharing summary removes direct identifiers but does not guarantee anonymity. Preview it before sharing.
- First rebuilding and ordinary source catch-up can still use substantial memory and temporary disk space. The aggregate cache limit is not a process-memory ceiling, and full replay or large detail requests remain expensive. Persisting aggregates also adds write work.
- Complete requests on the measured real corpus still exceeded 100 ms. Repeated compact requests can be faster, but the strict latency target remains unfinished.
- Native Codex plugin discovery and Hook registration require a compatible local Codex. Windows Hook collection remains unverified.
Verification and security
All five archives were built and exercised on their target platforms from commit 45712952c6545fbee76b6ac725b89dbaa1c9d11e. Installers and updates verify the published size and SHA-256 digest before switching versions. GitHub provides build provenance attestations for the platform archives. Wombat reads local sources without changing them; network behavior and native handoff follow the documented privacy boundaries.
Feedback and changes
Report problems in GitHub Issues. Private security reports should follow the security policy.
Changes since v0.3.1: compare v0.3.1...v0.3.2