pmem-ai 0.8.0-a → 0.8.0

package/README.md

@@ -29,7 +29,7 @@ The design is intentionally local and Git-friendly. Markdown cards remain the so
 - You want agents to update memory deliberately instead of auto-writing noisy logs.
 - You prefer local files over hosted memory services.
 
-It is not a vector database, MCP server, Graph UI, or remote multi-user service. v0.8 adds the Hybrid Recall Engine: deterministic multi-channel retrieval across exact IDs, aliases, tags, source file paths, always-on FTS5/BM25, and graph expansion — with recency scoring, stale/dirty penalties, and explainable output.
+It is not a vector database, MCP server, Graph UI, or remote multi-user service. v0.6 focuses on making the CLI agent-native with relationship discovery and polished workflows.
 
 ## Install
 
@@ -215,25 +215,21 @@ Do not edit SQLite directly. Edit Markdown cards or use pmem workflow commands,
 
 ### Recall And Ask
 
-Use `recall` for the current project state. v0.8 adds recall modes: `--mode brief` returns L0 + read-if-needed paths only, `--mode normal` (default) is the full agent context, and `--mode deep` preserves more detail when budget allows.
+Use `recall` for the current project state. v0.8 (upcoming on `main` branch, not yet released to npm) adds budget modes: `brief` returns L0 + read-if-needed paths, `normal` is the default agent context, and `deep` preserves more detail when budget allows.
 
 ```bash
 pmem recall --budget 2000
 pmem recall --mode brief --budget 500
-pmem recall --mode deep --budget 6000
 ```
 
-Use `ask` for targeted retrieval. v0.8 uses the Hybrid Recall Engine: a 5-stage deterministic pipeline — intent parse → multi-channel candidate generation → graph expansion → score fusion → L0-L3 budget packing. Channels include exact card ID, ID substring, exact title, title phrase, title token, aliases, tags, source file path, always-on FTS5/BM25, and graph hop expansion. Scores are multiplicative: `base × type_weight × recency × staleness_penalty × status`. Add `--explain` to see the `reasons[]` and `factors{}` for each matched card.
+Use `ask` for targeted retrieval. v0.8 (upcoming on `main` branch, not yet released to npm) uses the Hybrid Recall Engine: exact IDs, aliases, tags, source file paths, always-on FTS5/BM25, graph expansion, recency, and stale/dirty penalties are fused into a deterministic score. Add `--explain` to see why each card was recalled.
 
 ```bash
 pmem ask "sqlite runtime" --format compact
 pmem ask "src/core/query/recall.ts" --explain --limit 5
 pmem ask "release checklist" --format json
-pmem ask "module.core" --explain          # explain why this card was matched
 ```
 
-`--limit N` caps the result set (must be a positive integer; exit 2 on invalid value). `--explain` adds per-card `reasons` and `factors` fields to JSON output and a `Reasons:` section to compact output.
-
 ### Relations
 
 Use `relations` to inspect a card's edge graph and find pruning candidates:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pmem-ai",
-  "version": "0.8.0-a",
+  "version": "0.8.0",
   "description": "Project Memory for AI Agents: a local CLI runtime for project context, recall, and memory updates",
   "main": "dist/index.js",
   "bin": {

package/skills/pmem/SKILL.md

@@ -114,7 +114,7 @@ pmem related module.core --depth 2
 pmem trace decision.sqlite_runtime
 ```
 
-v0.8 adds recall modes to `pmem recall` and replaces `pmem ask` with the Hybrid Recall Engine: a 5-stage deterministic pipeline (intent parse → multi-channel candidate generation → graph expansion → score fusion → L0-L3 budget packing) that fuses exact IDs, aliases, tags, source file paths, FTS5/BM25, graph expansion, recency, and stale/dirty penalties into a ranked result. Use `--explain` to see per-card `reasons[]` and `factors{}`. The recall output is structured into:
+In v0.8 (upcoming on `main` branch, not yet released to npm), `pmem recall` remains trace-aware and adds recall modes, while `pmem ask` uses deterministic hybrid retrieval (exact IDs, aliases, tags, source files, FTS5/BM25, graph expansion, recency, and stale/dirty penalties). The recall output is structured into:
 - **PROJECT & STAGE**: The current metadata of the repository.
 - **CURRENT CONTEXT**: Recent summaries of what the project is doing.
 - **RECENT CHANGES**: Thick trace logs of what files and symbols changed.
@@ -317,7 +317,7 @@ pmem mcp
 | Tool | Description |
 |------|-------------|
 | `pmem_recall` | Restore project memory context (stage, focus, next, active cards, updates) |
-| `pmem_ask` | Search memory with Hybrid Recall Engine (exact ID/alias/tag/source-file → FTS5/BM25 → graph expansion → score fusion) |
+| `pmem_ask` | Search memory with 6-step retrieval (ID → alias → tag → graph → FTS5 → LIKE) |
 | `pmem_related` | Query graph neighbors of a card (edges grouped by type with direction/confidence) |
 | `pmem_status` | Detect changed files and affected memory cards (git or mtime) |