logosdb-memory
Semantic memory for Mistral Vibe — index code/docs, search across sessions, forget outdated knowledge using LogosDB + Mistral embeddings.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Semantic memory for Mistral Vibe — index code/docs, search across sessions, forget outdated knowledge using LogosDB + Mistral embeddings.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
| name | logosdb-memory |
| description | Semantic memory for Mistral Vibe — index code/docs, search across sessions, forget outdated knowledge using LogosDB + Mistral embeddings. |
| license | MIT |
| compatibility | Python 3.9+ |
| user-invocable | true |
| allowed-tools | ["bash","read_file","grep"] |
Gives Vibe persistent semantic memory across sessions by indexing your project into a local LogosDB vector store and performing sub-millisecond semantic search.
pip install 'logosdb[mistral]' # or: pip install logosdb requests numpy
export MISTRAL_API_KEY="..."
Add LOGOSDB_PATH to your environment (or .vibe/.env) to set where databases
are stored (default: ./.logosdb):
LOGOSDB_PATH=./.logosdb
MISTRAL_API_KEY=your_key_here
Three focused slash commands are available once this skill is enabled:
| Command | Example |
|---|---|
/ldb-index | /ldb-index ./src --namespace=backend |
/ldb-search | /ldb-search "JWT validation" --namespace=backend |
/ldb-forget | /ldb-forget --namespace=backend --id=42 |
$ cd myproject && vibe
> /ldb-index ./src --namespace=backend
Indexed 42 files into 'backend' collection
> Find where we handle JWT validation
Searching... Found 3 matches:
1. src/auth/jwt.ts (score: 0.94)
2. src/middleware/auth.ts (score: 0.87)
3. src/utils/token.ts (score: 0.72)
> Show me the first one
[Vibe displays src/auth/jwt.ts with explanation]
> /ldb-search "retry logic" --namespace=backend --top-k=3
Searching... Found 3 matches:
1. src/utils/retry.ts (score: 0.91)
2. src/api/client.ts (score: 0.83)
3. src/jobs/queue.ts (score: 0.77)
> /ldb-forget --namespace=backend --id=42
Deleted 1 entry from 'backend' namespace.
Natural-language queries (> Find where we...) also work without a slash command
— Vibe calls logosdb-vibe search automatically when it understands you are
asking about the indexed codebase.
logosdb-vibe index ./src --namespace code
logosdb-vibe search "retry logic" --namespace code --top-k 5
logosdb-vibe forget --namespace code --id 42
logosdb-vibe info --namespace code
logosdb-vibe list
from logosdb.vibe import VibeMemory
mem = VibeMemory(uri="./.logosdb") # reads MISTRAL_API_KEY from env
# Index an entire directory
mem.index("./src", namespace="code")
# Search
results = mem.search("where is JWT validated?", namespace="code", top_k=5)
for r in results:
print(r["score"], r["file"], r["text"][:120])
# Forget by semantic match
mem.forget(namespace="code", query="deprecated auth module")
# Forget by ID (from search result)
mem.forget(namespace="code", memory_id=42)
# Stats
print(mem.info())
Use separate namespaces to keep concerns isolated:
| Namespace | Content |
|---|---|
code | Source files (./src, ./lib) |
docs | Markdown, RST (./docs) |
tests | Test files (./tests) |
notes | Free-form session notes |
Vibe also supports MCP servers. To use the same LogosDB server as Claude Code,
see .vibe/config.toml in this project for the [[mcp_servers]] configuration.
Semantic memory management for Codex, backed by a local LogosDB MCP server. Handles explicit memory commands (/memory-recall, /memory-remember, /memory-index, /memory-status) and drives automatic silent indexing, searching, and turn-record storage. Invoke with: /memory-recall <query> | /memory-remember <text> | /memory-index | /memory-status
Delete an entry from LogosDB by row ID or by semantic query match.
Index a file or directory into LogosDB for semantic search across Vibe sessions.
Semantic search over an indexed LogosDB namespace.