Back to changelog

v1.3.0 · Independent Atomic Memory · Govern decision-model writes · Artifact merge and restoration

PowerContext 1.3.0 lets users manage each memory independently and adds policy controls, privacy protection, and traceable records when decision models assess which information is worth remembering. Users can also merge duplicate content, restore historical versions, and more easily find memories, Experiences, and Skills, alongside improved Server diagnostics and Agent integrations.

Highlights

  • Create, read, revise, search, forget, merge, and restore independent Atomic Memory Artifacts with immutable content history and explicit lifecycle state. Existing collection databases require the documented offline Atomic Memory migration before use.
  • Govern decision-model assessment of Memory write evidence with disabled, shadow, advisory, and enforcing modes. The gate remains opt-in; when enabled, it defaults to shadow and no_external_call, preserving writes without calling a model. Raw content evaluation requires explicit local_only configuration and a loopback backend. Scope-isolated durable observations retain policy versions, references, counts, and outcomes without candidate or evidence text; model failures preserve the write path. Enforcing promotion still requires calibration and evaluation; no public audit dashboard or hosted-redacted transport is included.
  • Share merge and restoration across Artifact families, preserving exact input revisions and historical evidence. Search Artifacts by family.
  • Add an unsigned Windows Desktop preview for connecting to an existing Server, selecting a Scope, saving notes, full-text search, reading exact revisions, and copying content and references. The preview also supports English/Chinese, themes, and connection diagnostics; complete native interaction and formal release qualification remain unfinished.
  • Choose automatic, full-text, semantic, or hybrid Memory search in the web Dashboard and inspect the executed mode, matching channels, and retrieval scores. Local Dashboard access can omit authentication; the Dashboard still requires explicit enablement.
  • Discover Server protocols and capabilities and inspect persistent deployment identity. Expose read-only extraction diagnostics and optional daily RecallEffort metrics; MCP request counts do not establish successful tool execution.
  • Import Codex sessions, refresh stale setup endpoints safely, bypass proxies for loopback Client transports, and improve DSH Desktop setup, native MCP catalog diagnostics, and handling of recoverable Server rejections.
  • Extend the Dify integration with 19 HTTP-backed tools and provide Experience and Skill applicability examples. Add compact Handoff continuity hints and improve ZCode lifecycle workflows and runtime diagnostics.
  • Persist smaller extraction windows after timeouts, stop redispatching exhausted Topic Memory windows, and classify deterministic rejections separately. Score recall sufficiency using vector relevance and keep SQLite vector searches within the supported result limit.
  • Expand paired continuation evaluation to Claude Code and Pi, task-completion workloads, and MemoryCode dialogues; report intervals, latency, and usage. These evaluation workflows do not imply a universal quality improvement.
  • Add a limited unified database migration bundle. The bundle covers four registered Artifact tables and explicitly reports server_ready=false; it is not a complete production Server upgrade path.
  • Upgrade the Server, clients, and Agent integrations together and reopen Agent sessions. Back up existing databases and follow the Atomic Memory migration guide; do not use the limited unified migration bundle as a replacement for the required migration.
uv tool install --force "powercontext[cli,server]==1.3.0"
View the full GitHub release