Back to changelog

v1.4.0 · Dream: evolve knowledge · Notes: retain progress · Traceable knowledge changes · Project memory for Kimi Code

PowerContext 1.4.0 lets users explicitly start Dream to propose knowledge revisions for review using task outcomes, corrections, and usage feedback, and retain current progress as Working Notes with evidence for Handoff. Browse and compare historical versions, and restore supported content as a new revision with its source and reason recorded. An experimental Kimi Code integration adds project memory saving, search, and relevant context injection in a new session.

Highlights

  • Dream: evolve knowledge. Use task outcomes, explicit corrections, or usage feedback as a reason to start Dream, selecting the relevant knowledge and evidence for a revision proposal. Propose changes to Atomic Memory, Profile, Topic Memory, Handoff, Prompt configuration, and tags through the unified Candidate review flow. Inspect the proposal and evidence before explicit approval. Available operations depend on Server configuration; Skill revision remains disabled until its trusted evaluation requirement is met. Handoff approval publishes a revision without executing or continuing work.
  • Notes: retain current progress. Capture findings, progress, and next steps as short-lived Working Notes, and cite still-valid evidence in a Handoff. Supported note kinds cover inspection, abandoned approaches, validation, unresolved questions, and host context. Notes retain exact Source identities and are excluded from automatic generation. Optional expiry prevents new consumption while preserving audit reads; physical cleanup and automatic promotion are not included.
  • Trace knowledge changes in the Dashboard: browse revision history, compare exact versions, and roll supported historical content forward as a new revision with the source, reason, and actor recorded. Atomic Memory history and comparison are read-only here; restoration uses its separate preview and confirmation flow. New revision timestamps do not invent dates for older records.
  • Connect Kimi Code to project memory through an experimental integration: save and search memories, read exact revisions, and receive relevant context in a new session. Use an explicitly configured Scope, with setup, diagnostics, native MCP operations, prompt capture, and content-free Hook runtime observations.
  • Opt in to bounded startup context in Codex and Claude Code. Select active Atomic Memories with the bootstrap-context tag and optionally supply an exact committed Handoff. Delivery receipts support lifecycle deduplication and keep fully delivered exact Memory revisions out of the first ordinary recall; content remains cited, untrusted history and the feature is disabled by default.
  • Extend Atomic Memory retrieval through the shared Artifact search path, unify Chinese segmentation and English word analysis across supported backends, and configure full-text/vector fusion weights for Atomic Memory and Topic Memory. Preserve supporting Source language during generation, reduce answer-format noise in text queries, and handle long SQLite and OceanBase queries.
  • Expand DSH to native MCP workflows with catalog checks, Scope enforcement, approval handling, and caller-content secret screening. Refresh generated host guidance while preserving locally edited guidance.
  • Add scope-local stable handles that permanently name an exact Artifact revision, alongside revision metadata for tracing replacements and restorations. Binding a handle requires Scope administration; resolving it preserves the referenced immutable version.
  • Add bounded request sanitization for decision backends, including structured subject and evidence values, known sensitive values, and supported provider-key formats. Projection changes do not rewrite authoritative content or decide write policy, and the sanitizer is not a general PII detector. The legacy collection MemoryWriteGate remains unsupported by the Atomic Runtime.
  • Standardize structured operational logs, metrics, and trace identity, expose extraction diagnostics in CLI capabilities, preserve exact captured Source identities on reads, and pin SQLite content before custom authorization. Improve cancellation cleanup and omit the placeholder x-api-key when Anthropic-compatible requests use an explicitly configured Authorization header.
  • Add cross-host work-continuity acceptance, offline Memory write-gate evaluation, and Dream generation baselines. Extend Kimi Code and ZCode acceptance workflows and refresh Desktop contract validation. These workflows do not establish a universal quality gain or change Windows Desktop preview qualification.
  • Upgrade Server, Client, CLI, MCP, and Agent integrations together. Legacy Memory collections and citation models leave online APIs; retained Memory routes return Atomic Memory records, and unsupported collection operations fail explicitly. Back up data, stop writers, and complete offline Atomic Memory conversion and verification. Candidate APIs move from /v1/artifact-candidates/* to /v1/candidates/* without aliases, and Dream requires contract 2. Existing Dream storage needs coordinated migration after draining unfinished runs; the current unified database bundle still reports server_ready=false and does not provide a complete production upgrade path. Do not bypass migration_required or migration_framework_not_ready.
uv tool install --force "powercontext[cli,server]==1.4.0"
View the full GitHub release