v1.4.1 · Faster recall and Memory writes · Portable archives · More reliable Agent and Desktop workflows
PowerContext 1.4.1 reduces repeated database work during recall and Memory writes, removes duplicate Source text from extraction requests, and adds offline archive export and restore. It also expands Kimi Code workflows, improves Desktop navigation and save recovery, and adds dependency and processing observations alongside managed logs.
Highlights
- Reduce repeated database work by consolidating context recall queries, batching Memory lifecycle and authorization checks, reusing exact evidence revisions, and batching cited Experience reads. Scheduled processing skips consumed Scope windows, and Source capture can proceed while Memory extraction is running.
- Remove duplicate Source text from Atomic Memory extraction requests and compact repeated bodies in additional model projections. Skip reconciliation when a candidate has no comparable memories, and return invalid reconciliation decisions to the model for correction within the configured retry budget.
- Export, inspect, validate, and restore selected complete Scopes as portable .pcb archives. Preserve exact identities, revisions, evidence, and supported review state; validate conflicts before writes and recognize identical records on replay. This is an offline operator workflow, excludes deployment credentials and derived indexes, and does not replace a complete database backup or schema migration.
- Complete the Server migration framework through db-migrate status, plan, apply and verify for registered Server storage. Exact source layouts, reviewed plans, maintenance locks and explicit backup policies guard schema and data transitions; incomplete validation cannot report full Server readiness.
- Preserve registered source history and original recovery evidence across interrupted upgrades and idempotent retries. Startup validates authoritative records, search projections, embedding profiles and retained history before serving requests.
- Display migration steps as v1_4_1_001 through v1_4_1_003 while preserving existing revision IDs and checksums. The 1.3.0 and 1.4.0 source bridges remain separate; registration does not certify every intermediate software release.
- Declare twelve 1.3.0 upgrade combinations across SQLite, embedded seekDB, and OceanBase, with vectors and Code enabled or disabled. Publication requires the ten automatic cases to pass against the exact target artifact. The two seekDB cases with Code enabled remain manual-only, unverified, and excluded from supported upgrades; the 1.4.0 bridge is outside this CI matrix. Failed, skipped, or missing automatic cases block publication, and support remains upgrade-required until acceptance and the reviewed support declaration are complete.
- Inspect durable Topic Memory processing blocks and explicitly resume eligible work after correcting its cause. Reject Topic Memory semantic mismatches, make the recall embedding budget configurable, and report why preparation fell back from semantic recall.
- Expand the experimental Kimi Code integration with workspace Scope bindings and opt-in recommended/full tool profiles, including Handoff, Experience, managed Skill, and Candidate workflows. Initial setup still defaults to the minimal profile; larger profiles require the corresponding Server capabilities and explicit approval for review operations.
- Improve Desktop shared navigation, Scope recovery, exact revision reading, and draft preservation after failed or ambiguous saves. Add an unsigned macOS app and disk-image preview and fix Linux packaged navigation. Protocol checks and hosted-runner tests do not establish signed distribution, clean-machine installation, or complete native accessibility coverage.
- Add an independent local Ops tool for inspecting, configuring, and managing a prepared personal Server runtime through explicit preflight and apply operations. Runtime downloading, repair, and the Desktop installation bridge remain separate work. Add a Scope CLI command group and text search for Experiences in the web Dashboard.
- Add managed rotating log files and worker log aggregation, plus dependency and processing observations. Bound HTTP request bodies and active responses, authenticate MCP calls per HTTP request, preserve path authorization, and make local write ownership atomic. Repair eligible ownerless migrated memories without inventing historical owners.
- Aggregate streamed model output so generation can survive gateway idle timeouts while retaining a complete result. Python applications can explicitly inject the optional Jev reranker with its own credentials and pinned model; it is not enabled by default or configured through the Server environment settings.
- Guard automatic prompt capture and context preparation, bound long-prompt recall and WorkBuddy Scope discovery, and improve Codex shim resolution and Windows receipt replacement. Reject blank text Sources at creation so they cannot poison later processing; make tool failures identify the rejected field or reason.
- Upgrade Server, clients, and Agent integrations together only when the source version and backend have an accepted path for the exact target artifact. Stop writers, back up data and configuration, and review the migration plan. Drain unsupported queued or running 1.3.0 Dream tasks, preserve the historical embedding configuration, and follow the documented recovery procedure after interruption. Complete migration and full Server readiness verification before restarting services.
uv tool install --force "powercontext[cli,server]==1.4.1"View the full GitHub release