LLM内存控制服务

pip install memctl

pipx install memctl

pip install memctl[mcp]          # pip
pipx install "memctl[mcp]"      # pipx (recommended for CLI use)

pip install memctl[docs]
pipx inject memctl python-docx python-pptx openpyxl odfpy pypdf   # pipx

pip install memctl[all]
pipx install "memctl[all]"      # pipx

memctl init
# Creates .memory/memory.db, .memory/config.json, .memory/.gitignore

eval $(memctl init)
# Sets MEMCTL_DB=.memory/memory.db

# Ingest source files + recall matching items → injection block on stdout
memctl push "authentication flow" --source src/auth/

# Ingest Office documents (requires memctl[docs])
memctl push "project status" --source reports/*.docx slides/*.pptx

# Ingest PDFs (requires pdftotext)
memctl push "specifications" --source specs/*.pdf

# Recall only (no ingestion)
memctl push "database schema"

# Pipe LLM output into memory
echo "We chose JWT for stateless auth" | memctl pull --tags auth,decision --title "Auth decision"

# Or pipe from any LLM CLI
memctl push "API design" | llm "Analyze this" | memctl pull --tags api

# Human-readable
memctl search "authentication"

# JSON for scripts
memctl search "database" --json -k 5

# Auto-mounts, auto-syncs, and inspects — all in one command
memctl inspect docs/

# Same in JSON (for scripts)
memctl inspect docs/ --json

# Skip sync (use cached state)
memctl inspect docs/ --no-sync

# One-shot: auto-mount, auto-sync, inspect + recall → LLM → answer
memctl ask docs/ "What authentication risks exist?" --llm "claude -p"

# With Ollama
memctl ask src/ "What is under-documented?" --llm "ollama run granite3.1:2b"

# JSON output with metadata
memctl ask docs/ "Summarize the architecture" --llm "claude -p" --json

# Interactive chat with any LLM
memctl chat --llm "claude -p" --session

# With pre-ingested files and answer storage
memctl chat --llm "ollama run granite3.1:2b" --source docs/ --store --session

memctl show MEM-abc123def456     # Show item details
memctl stats                     # Store metrics
memctl stats --json              # Machine-readable stats
memctl consolidate               # Merge similar STM items
memctl consolidate --dry-run     # Preview without writing

memctl <command> [options]

memctl init [PATH] [--force] [--fts-tokenizer fr|en|raw]

memctl push QUERY [--source FILE ...] [--budget N] [--tier TIER] [--tags T] [--scope S]

echo "..." | memctl pull [--tags T] [--title T] [--scope S]

memctl search QUERY [--tier TIER] [--type TYPE] [-k N] [--json]

memctl consolidate [--scope S] [--dry-run] [--json]

memctl push "question" | memctl loop "question" --llm "claude -p" [--max-calls 3] [--protocol json]

# Iterative recall with Claude
memctl push "How does authentication work?" --source docs/ \
  | memctl loop "How does authentication work?" --llm "claude -p" --trace

# Sovereign local LLM
memctl push "database schema" --source src/ \
  | memctl loop "database schema" --llm "ollama run granite3.1:2b" --protocol json

# Replay a trace (no LLM needed)
memctl loop --replay trace.jsonl "original question"

memctl mount PATH [--name NAME] [--ignore PATTERN ...] [--lang HINT]
memctl mount --list
memctl mount --remove ID_OR_NAME

memctl sync [PATH] [--full] [--json] [--quiet]

# Orchestration mode — auto-mounts, auto-syncs, and inspects
memctl inspect PATH [--sync auto|always|never] [--no-sync] [--mount-mode persist|ephemeral]
                    [--budget N] [--ignore PATTERN ...] [--json] [--quiet]

# Classic mode — inspect an existing mount by ID/name
memctl inspect --mount ID_OR_NAME [--budget N] [--json] [--quiet]

memctl ask PATH "question" --llm CMD [--inspect-cap N] [--budget N]
           [--sync auto|always|never] [--no-sync] [--mount-mode persist|ephemeral]
           [--protocol passive|json|regex] [--max-calls N] [--json] [--quiet]

memctl chat --llm CMD [--session] [--store] [--folder PATH]
            [--protocol passive|json|regex] [--max-calls N] [--budget N]
            [--source FILE ...] [--quiet]

memctl export [--tier T] [--type T] [--scope S] [--include-archived]

# Export all items
memctl export > backup.jsonl

# Export only LTM decisions
memctl export --tier ltm --type decision > decisions.jsonl

# Pipe between databases
memctl export --db project-a.db | memctl import --db project-b.db

memctl import [FILE] [--preserve-ids] [--dry-run]

# Import from file
memctl import backup.jsonl --db fresh.db

# Dry run — see what would happen
memctl import backup.jsonl --dry-run

# Preserve original IDs (for controlled migration)
memctl import backup.jsonl --preserve-ids --db replica.db

{
  "store": {"fts_tokenizer": "fr"},
  "inspect": {
    "dominance_frac": 0.40,
    "low_density_threshold": 0.10,
    "ext_concentration_frac": 0.75,
    "sparse_threshold": 1
  },
  "chat": {"history_max": 1000}
}

export MEMCTL_DB=.memory/memory.db

# Shortcuts
meminit()  { memctl init "${1:-.memory}"; }
memq()     { memctl push "$1"; }                        # recall only
memp()     { memctl push "$1" ${2:+--source "$2"}; }    # push with optional source
mempull()  { memctl pull --tags "${1:-}" ${2:+--title "$2"}; }

# Ingest docs + recall + feed to LLM + store output
memctl push "API design" --source docs/ | llm "Summarize" | memctl pull --tags api

# Search and pipe to jq
memctl search "auth" --json | jq '.[].title'

# Batch ingest a directory
memctl push "project overview" --source src/ tests/ docs/ -q

# Export all items as JSONL backup
memctl export > backup.jsonl

# Export only LTM items
memctl export --tier ltm > decisions.jsonl

# Import into a fresh database
memctl import backup.jsonl --db fresh.db

# Pipe between databases
memctl export --db project-a.db | memctl import --db project-b.db

# Dry-run import to check counts
memctl import backup.jsonl --dry-run

# Iterative recall-answer loop with trace
memctl push "auth flow" --source docs/ | memctl loop "auth flow" --llm "claude -p" --trace

# One-liner: inspect a folder (auto-mount + auto-sync)
memctl inspect docs/

# Inspect in JSON, pipe to jq for extension breakdown
memctl inspect src/ --json | jq '.extensions'

# Inspect without syncing (use cached state)
memctl inspect docs/ --no-sync --json

# One-shot folder Q&A (inspect + scoped recall + LLM)
memctl ask docs/ "What are the auth risks?" --llm "claude -p"

# Folder Q&A with JSON output
memctl ask src/ "Summarize the architecture" --llm "claude -p" --json

# Interactive folder-scoped chat
memctl chat --llm "claude -p" --folder docs/ --session --store

# Interactive chat with pre-ingested docs
memctl chat --llm "claude -p" --source docs/ --session --store

# Claude Code (default) — cross-platform (Python)
memctl setup mcp

# Claude Desktop
memctl setup mcp --client claude-desktop

# Both clients, non-interactive
memctl setup mcp --client all --yes

# Custom database path
memctl setup mcp --db ~/my-project/.memory/memory.db

# Preview without changes
memctl setup mcp --dry-run

# 1. Install
pip install "memctl[mcp]"

# 2. Initialize workspace
memctl init ~/.local/share/memctl

# 3. Verify
memctl serve --check --db ~/.local/share/memctl/memory.db

{
  "mcpServers": {
    "memctl": {
      "command": "memctl",
      "args": ["serve", "--db", "~/.local/share/memctl/memory.db"]
    }
  }
}

{
  "mcpServers": {
    "memctl": {
      "command": "memctl",
      "args": ["serve", "--db", "~/.local/share/memctl/memory.db"]
    }
  }
}

memctl serve --db ~/.local/share/memctl/memory.db
# or
python -m memctl.mcp.server --db ~/.local/share/memctl/memory.db

# Default: db-root enforced, rate limits on, audit to stderr
memctl serve --db project/memory.db

# Explicit secure mode with audit file
memctl serve --db memory.db --db-root . --audit-log audit.jsonl

# Disable rate limits (development only)
memctl serve --db memory.db --no-rate-limit

# Install safety guard + audit logger hooks (cross-platform)
memctl setup hooks

# Uninstall
memctl teardown hooks

pip install "memctl[mcp]"
memctl setup eco --db-root .memory
memctl eco on    # Enable eco mode (required)

memctl teardown eco
# Removes hook + strategy file. Preserves .memory/memory.db and MCP config.

memctl/
├── types.py           Data model (MemoryItem, MemoryProposal, MemoryEvent, MemoryLink)
├── store.py           SQLite + FTS5 + WAL backend (10 tables + schema_meta)
├── extract.py         Text extraction (text files + binary format dispatch)
├── ingest.py          Paragraph chunking, SHA-256 dedup, source resolution
├── policy.py          Write governance (35 patterns: secrets, injection, instructional, PII)
├── config.py          Dataclass configuration + JSON config loading
├── similarity.py      Stdlib text similarity (Jaccard + SequenceMatcher)
├── loop.py            Bounded recall-answer loop controller
├── mount.py           Folder mount registration and management
├── sync.py            Delta sync with 3-tier change detection
├── inspect.py         Structural inspection and orchestration
├── chat.py            Interactive chat REPL (readline history, multi-line)
├── ask.py             One-shot folder Q&A orchestrator
├── query.py           FTS query normalization and intent classification
├── export_import.py   JSONL export/import with policy enforcement
├── cli.py             28 CLI commands
├── installer.py       Cross-platform setup/teardown (MCP, eco, hooks)
├── consolidate.py     Deterministic merge (Jaccard clustering, no LLM)
├── proposer.py        LLM output parsing (delimiter + regex + JSON stdin)
└── mcp/
    ├── tools.py       21 MCP tools (memory_* prefix)
    ├── formatting.py  Injection block format (format_version=1)
    └── server.py      FastMCP server entry point

git clone git@github.com:ovitrac/RAGIX.git
cd RAGIX
pip install -e .[all]
# Point at the same database — all items carry over
ragix memory stats --db /path/to/your/.memory/memory.db

from memctl import MemoryStore, MemoryItem, MemoryPolicy

# Open or create a store
store = MemoryStore(db_path=".memory/memory.db")

# Write an item
item = MemoryItem(
    title="Architecture decision",
    content="We chose event sourcing for state management",
    tier="stm",
    type="decision",
    tags=["architecture", "event-sourcing"],
)
store.write_item(item, reason="manual")

# Search
results = store.search_fulltext("event sourcing", limit=10)
for r in results:
    print(f"[{r.tier}] {r.title}: {r.content[:80]}")

# Policy check
policy = MemoryPolicy()
from memctl.types import MemoryProposal
proposal = MemoryProposal(
    title="Config", content="Some content",
    why_store="Important finding",
    provenance_hint={"source_kind": "doc", "source_id": "design.md"},
)
verdict = policy.evaluate_proposal(proposal)
print(verdict.action)  # "accept", "quarantine", or "reject"

store.close()

pip install memctl[dev]
pytest tests/ -v