This release adds 3 notable features for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Topics
+2 more
Summary
AI summaryAlluvia now runs concurrently with WAL mode and adds status/diagnostic commands.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Feature | Medium |
Runs concurrent sessions using WAL mode; readers never block refresh. Runs concurrent sessions using WAL mode; readers never block refresh. Source: llm_adapter@2026-07-18 Confidence: high |
— |
| Feature | Medium |
`alluvia status` reports all paths, sizes, and data class breakdown; supports `--json`. `alluvia status` reports all paths, sizes, and data class breakdown; supports `--json`. Source: llm_adapter@2026-07-18 Confidence: high |
— |
| Feature | Medium |
`alluvia doctor` diagnoses and automatically applies safe repairs (WAL upgrade, schema migrations, pruning, permission fixes). `alluvia doctor` diagnoses and automatically applies safe repairs (WAL upgrade, schema migrations, pruning, permission fixes). Source: llm_adapter@2026-07-18 Confidence: high |
— |
| Feature | Medium |
`--verbose` and `refresh --plan` provide pipeline preview without LLM usage. `--verbose` and `refresh --plan` provide pipeline preview without LLM usage. Source: llm_adapter@2026-07-18 Confidence: high |
— |
| Bugfix | High |
Any alluvia process can be killed at any instant with zero corruption and no cleanup debt; SIGTERM pauses refresh safely. Any alluvia process can be killed at any instant with zero corruption and no cleanup debt; SIGTERM pauses refresh safely. Source: llm_adapter@2026-07-18 Confidence: high |
— |
Full changelog
The good-citizen release: alluvia now runs concurrently, dies cleanly, shows
you everything it keeps, and repairs itself (#5, #6, #7, #8).
- Concurrent sessions (#5) — the store runs in WAL mode (readers never
block a running refresh; legacy stores upgrade transparently), refresh
takes a single-writer lock released by the OS on any process death
("already running (pid N)" instead of double-spending your LLM budget),
andservereuses an already-running dashboard or walks to a free port. - Kill-anytime contract (#5) — any alluvia process can be killed at any
instant with zero corruption and zero cleanup debt. SIGTERM behaves like
Ctrl-C; interrupting a refresh prints "paused — everything done so far is
saved" and resumes on the next run. alluvia status(#6) — every path alluvia touches with sizes, the
store broken down by data class (raw = source of truth · derived =
rebuildable · judgments = yours), and what's live right now.--jsonfor
scripting.alluvia doctor(#7) — diagnoses the whole installation and applies
every safe repair automatically: WAL on legacy stores, schema migrations,
pruning orphaned derived rows, config permissions, stale digest flags,
impossible governor cooldowns. Raw data and judgments are never touched.
--checkreports without repairing (exit 1 if repairs are needed),
--liveproves your provider key with one tiny call, and
--rebuild-derivedis the confirmed recovery lever that discards derived
data while raw sessions and your ratings survive.--verboseandrefresh --plan(#8) — see the pipeline and governor
think; preview exactly what a refresh would do (sessions pending, theme
work, cooldowns in effect) without spending a single LLM call.
Install: uv tool install alluvia · PyPI · Closes #5, #6, #7, #8
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About Alluvia
All releases →Related context
Related tools
Beta — feedback welcome: [email protected]