CPA Manager Plus: baixar versions anteriors

Overview
v1.14.9 is a 1.x security-maintenance release after v1.14.8. It contains the merged dependency and CI-security maintenance changes in #1049. The focus is to use patched build toolchains and dependencies and prevent future changes with failing dependency scans from passing the existing CI gate.
There are no new user-facing features. This release does not include v2 architecture work or change CPA management APIs, authentication, database schemas, credential/configuration formats, or historical usage data.
Security and Build Maintenance
- Go and Manager Server (#1049): Require Go 1.26.10 for source builds, update golang.org/x/sys to v0.44.0, select the CI Go version from go.mod, and pin the Docker Go builder. The SQLite dependency remains at v1.46.1.
- Node.js and frontend dependencies (#1049): Pin Docker's frontend builder to Node.js 22.23.4 and CI to 24.22.1. Update axios, React Router, Vite, Vitest, and related locked dependencies.
- Documentation dependencies (#1049): Resolve relevant VitePress/DocSearch advisories with scoped dependency overrides while retaining VitePress 1.6.4, without undertaking a documentation-site redesign.
- Ongoing dependency security (#1049): Add a full-workspace npm audit and Go vulnerability scans for Linux, macOS, and Windows on both amd64 and arm64. Missing, skipped, or failed results now block Required checks. Add scheduled security scans and weekly Dependabot updates targeting dev.
- The verification recorded on #1049 reduced the npm audit from 26 affected package nodes to zero reported advisories and found no known issues in the six Go source scan targets. These are findings against the vulnerability database snapshot used for that verification, not a guarantee against future advisories.
Upgrade Notes
- Upgrade from v1.14.8 and restart normally. No database migration, configuration rewrite, or credential recreation is required; migration_required remains false.
- Users of prebuilt Docker images or native packages do not need to install Go or Node.js on the host. Source builds of Manager Server now require Go 1.26.10; Node.js builds require compatible 22.23.4+ (22.x), 24.22.1+ (24.x), or 26.12.1+ runtimes.
- The dependency/toolchain changes are substantial, but no user-visible UI, authentication, API, or data behavior changes are intended. Check the management UI, existing connections, and key workflows after upgrading.
- The CPA compatibility baseline for full Muse / Meta features remains v7.3.4+.
- The existing offline derived-index maintenance guidance from v1.14.7 still applies; this release adds no further maintenance steps.
- #1049's local verification covered frontend/server tests, offline pages, and native cross-builds. Local Docker build completion was blocked by a base-image metadata fetch error. Formal CI and the release Dry Run remain the authority for publish readiness.

Full Changelog: https://github.com/seakee/CPA-Manager-Plus/compare/v1.14.8...v1.14.9

Overview
v1.14.8 is a 1.x maintenance release after v1.14.7. It includes four merged frontend fixes: background-refresh data continuity (#1001), provider-aware source attribution in request monitoring (#1014), quota-window presentation (#1020), and xAI OAuth subscription-plan and observed-quota handling (#1006).
This release contains no v2 architecture work and does not change the Manager Server, CPA management APIs, database schema, credential format, or configuration format. There are no new user settings or migration steps.
Fixes
- Background-refresh continuity (#1001): Keep loaded credential history, quota usage, Usage Analytics data, and model-price rows visible while refreshing within the same connection and data scope, including temporary request failures. Avoid flashing, blank content, or false empty states.
- Credential replacement invalidates historical and quota-usage evidence belonging to the previous credential identity without discarding unaffected rows. Rolling-range day changes and changed filters still invalidate out-of-scope data.
- Request-monitoring source attribution (#1014, fixes #1013): Disambiguate shared API keys and host ordinals across providers using the event provider where available. Preserve configured upstream labels; fall back to an unattributed source rather than guessing when the provider is missing or candidates remain ambiguous. Raw API keys are not displayed.
- Quota-window presentation (#1020): Show up to five principal Antigravity quota windows in both cards and table rows, including a separate Gemini 4 Argon group. Distinguish confirmed empty model usage from unknown, stale, partial, or unavailable quota data.
- xAI OAuth subscriptions (#1006, fixes #1005): When a CPA auth file lacks plan information, resolve the subscription in the background through CPA's existing api-call endpoint. Display a confirmed SuperGrok Heavy plan where available, and continue showing observed quota even when subscription status cannot be confirmed.
- Keep xAI quota observations separate from account-health entitlement decisions: stale plan evidence cannot establish availability or cross connection/credential boundaries. Subscription lookups are deduplicated through a low-priority serial lane after quota refresh; no periodic polling is added.
Upgrade Notes
- Upgrade from v1.14.7 and restart normally. No database migration, configuration rewrite, or credential recreation is needed; migration_required remains false.
- The subscription lookup in #1006 uses the existing CPA api-call capability and upstream subscription endpoints. If these are unavailable, the plan can remain unknown while observed quota stays visible; old plan labels alone do not establish availability.
- The retained snapshots in #1001 are scoped to the same connection and credential/filter identity. Scope changes still invalidate previous evidence.
- The minimum CPA baseline for full Muse / Meta functionality remains v7.3.4+; this release does not raise it.
- The existing v1.14.7 offline-maintenance guidance for the latency-analytics covering index on populated databases still applies. This release introduces no further indexes or maintenance commands.
Acknowledgements
- @M1nt-Ch0c0 - Fixed xAI OAuth subscription-plan recognition with background lookup and preserved observed quota when the plan is unknown (#1006).
- @imrooki - Fixed missing Antigravity quota windows and clarified confirmed-empty versus unknown model-usage states (#1020).

Full Changelog: https://github.com/seakee/CPA-Manager-Plus/compare/v1.14.7...v1.14.8

Overview
v1.14.7 is a 1.x maintenance release after v1.14.6. It includes five already-merged fixes and performance improvements: a SQLite covering index for long-window latency analytics (#960), removal of history-sized candidate reads from the quota-snapshot writer hot path (#981), the Antigravity Hub client-identity compatibility fix (#986), unified non-blocking background refresh behavior across v1 views (#987), and inclusion of the first request after a Codex five-hour rollover (#991).
This release adds no new 1.x feature, does not change CPA management APIs, credential privileges, the SQLite schema, or configuration formats, and introduces no new polling jobs or timers.
Highlights
Fixes
- Background refresh behavior (#987): Loaded credentials, configuration, OAuth rules, Dashboard data, logs, plugins, System data, provider model discovery, Usage Analytics, and action-candidate views now retain the current successful snapshot while revalidating within the same data scope instead of reverting to full loading, blank content, or a false empty state.
- A transient revalidation failure keeps the last successful content visible with lightweight failure feedback, while a successful empty response is treated as a loaded empty snapshot. Connection, provider, credential, and capability scope changes invalidate stale data, late responses, and captured destructive actions.
- Loaded credential configuration remains visible but read-only after a reload failure and becomes editable again after a successful retry. Log reading position, plugin iframes, and loaded charts remain stable during ordinary refreshes.
- Antigravity quota compatibility (#986): Replace the obsolete antigravity/cli/1.0.13 management-request identity with the Hub identity used by current CPA so Cloud Code no longer truncates retrieveUserQuotaSummary because of the old-client floor. Gemini weekly and five-hour quota buckets are available again without changing endpoints, polling, or quota parsing semantics.
- First Codex request after a five-hour rollover (#991): When the previous cycle has ended but a request starts a few seconds before the response headers establish the new scheduled cycle start, that confirmed request now belongs to the new cycle's actual interval rather than being omitted from both cycles. Scheduled boundaries remain unchanged and the request is not double counted.
- Quota Snapshot write performance (#981): Remove the history-sized candidate query and sort from the writer hot path while leaving cross-request boundary canonicalization in the existing lifecycle repository. External quota APIs, displayed semantics, and persistence formats are unchanged.
Performance
- Long-window latency analytics (#960): Add a (timestamp_ms, latency_ms, ttft_ms) covering index through the existing deferred/offline derived-index lifecycle while retaining production SQL, the timestamp index, and exact percentile definitions.
- On a warmed synthetic 3.2M-event database, the 30-day raw SQL scan median fell from about 3.77 seconds to 0.43 seconds. The exact 120-day percentile SQL remains sorting-bound, improving from roughly 49 seconds to 25 seconds in that experiment. These figures describe the measured index boundary and are not an end-to-end guarantee for every deployment.
- The index adds modest disk and write overhead. Complex provider/model/account filters can still require table reads, and very long exact-percentile windows remain a future optimization area; this maintenance release does not expand into pre-aggregation or percentile-algorithm redesign.
Upgrade Notes
- A normal upgrade and restart is sufficient for all correctness fixes in v1.14.7. There is no SQLite schema or configuration-format migration, so migration_required remains false.
- On an existing non-empty usage.sqlite, the new latency covering index from #960 follows the existing deferred/offline maintenance lifecycle. Skipping that maintenance does not prevent Manager Server from starting or serving 

Overview
v1.14.6 is a maintenance patch after v1.14.5, limited to three already-merged fixes: Codex workspace spend-control limits (#935 / #938), retained monitoring data protection during usage archive cleanup (#939 / #945), and cache-token double counting for otherwise-unknown events with strict CPA v2 accounting evidence (#941 / #946).
This release does not introduce new features, polling, or timers, and it does not change CPA management APIs, credential privileges, or the SQLite schema.
Highlights
Fixes
- Codex workspace spend control (#935 / #938): Correctly interpret structured spend_control.individual_limit while retaining legacy scalar and persisted browser quota-cache compatibility. Preserve budget, used/remaining amounts, units, time windows, and reset evidence. An exhausted spend-control limit remains a limited/quota-risk state, even if standard 5-hour or weekly quotas have headroom.
- Explain zero-budget 0 / 0 cases as a workspace-budget restriction rather than a misleading generic “100% used” limit. Suggest reviewing the workspace budget only when complete spend-control evidence is the sole limitation; preserve the existing quota-refresh recommendation when other quota signals compete. No background provider probing or polling is added.
- Archive cleanup integrity (#939 / #945): Before each bounded raw-event deletion batch, verify that the corresponding retained Monitoring projection rows exist and have matching timestamps. Repair only missing/mismatched rows from still-live raw data and verify the handoff within the same SQLite transaction; fail closed and roll back if integrity cannot be established.
- On cleanup resume, verify retained projections for events deleted earlier in the same run before allowing more deletion. This prevents future losses of this kind but does not reconstruct already permanently deleted raw history.
- CPA v2 canonical cache accounting (#941 / #946): For previously unclassified events, use canonical cache semantics only when CPA v2 accounting_version=2 and token_breakdown.schema_version=2 provide a complete, internally consistent breakdown across input, cache, output, reasoning, and total tokens. This prevents counting cache-read tokens twice.
- Preserve existing explicit and known provider/executor/model precedence; incomplete, ambiguous, unknown-version or inconsistent evidence keeps the previous fallback. Standard JSONL imports/exports distinguish upstream-declared cache modes from CPAMP-derived accounting hints rather than promoting derived values into raw upstream evidence.
- Automatically reconcile eligible historical online usage_events through the existing accounting-semantics migration path and rebuild affected usage/pricing aggregates. No manual SQLite edits or offline cleanup command are needed.
Upgrade Notes
- A normal upgrade and restart is sufficient. There is no SQLite schema or configuration format migration, but eligible historical events undergo an automatic accounting-semantics correction and corresponding usage/cost rollups may change.
- Keep a consistent database and archive backup before upgrading. If exact pre-upgrade historical accounting must be restored after the automatic rewrite, restore the matching pre-upgrade database/archive backup instead of manually reversing derived fields.
- Strict canonical cache accounting is used only when complete CPA v2 evidence exists; other events preserve previous compatibility behavior. The minimum CPA baseline for full Muse / Meta functionality remains v7.3.4+.
- Archive deletion may incur a small bounded verification cost; cleanup fails closed if retained material coverage cannot be proven.
- No new background polling, timers, or manual maintenance commands are introduced.

Full Changelog: https://github.com/seakee/CPA-Manager-Plus/compare/v1.14.5...v1.14.6

Overview
v1.12.6 focuses on secure CPA connection migration and rollback, provider-aware plan and quota presentation, safe Codex history continuity, and quota-window recovery across observation gaps. Manager Server keeps CPA Management Keys encrypted at rest and out of browser responses; installer upgrades preserve recoverable state; Accounts and Request Monitoring share current provider quota evidence; and historical reads fail closed when identity cannot be proved.
Highlights
Fixes
- Manager Server configuration responses now expose only whether the CPA Management Key is configured; saved keys remain encrypted server-side, while setup and rotation use write-only input.
- Docker and native installers import legacy CPA URL/key inputs into encrypted SQLite, validate health, administrator authentication, and CPA management access before removing legacy runtime wiring, and preserve external key files.
- Migration snapshots and restores treat usage.sqlite, WAL/SHM/journal companions, and data.key as one complete file set; cancellation and failures retain a safe rollback path.
- Legacy CPA connections follow explicit authority rules: complete manager_config_v1 wins, compatible setup data can repair a partial manager record, and conflicting or unresolvable state requires store-cpa-connection --repair-conflict instead of silent rebinding.
- Failed CPA imports retain a permission-restricted pending state and temporary key for safe retry, while malformed post-v2 persisted keys and missing or mismatched data.key fail closed without replacement-key creation.
- Accounts and Request Monitoring now share provider quota refresh results and prevent stale cache generations or response-header evidence from overwriting newer quota state.
- Codex account history and Monitoring account windows reuse legacy data only when explicit account identity and exact credential provenance match; ambiguous or foreign records remain separated, and usage_events stays immutable.
- Accounts quota details can show the most recently closed 5-hour or weekly window after a natural rollover followed by an observation gap.
- Request Monitoring plan filters preserve provider-scoped identity, so equivalent aliases aggregate within one provider without colliding with another provider's plans.
Refactor
- Accounts, Monitoring, Inspection, Dashboard, and Usage Analytics use one provider-aware plan presentation path with compact/full labels, raw plan preservation, and visible fallbacks for unknown values.
Docs
- Deployment and backup documentation now describe encrypted CPA storage, complete snapshot files, retry and conflict-repair procedures, and rollback cleanup boundaries.
Upgrade Notes
- Existing deployments may run the connection-storage normalization migration during upgrade startup; keep usage.sqlite, WAL/SHM/journal files, and the matching data.key together in backups. The migration does not rewrite usage_events.
- Valid persisted connections need no manual configuration change. If an old env/secret import fails, preserve the installer-managed pending file and temporary key for retry; external CPA_MANAGEMENT_KEY_FILE files are not deleted.
- If the service reports corrupted post-v2 CPA connection storage or an unresolved conflict, do not replace data.key; stop Manager Server and use store-cpa-connection --repair-conflict with the explicitly verified URL and key file.
- Monitoring integrations should use opaque row.id for state and row.account for display, and must not merge Codex accounts by project ID or display value alone.