Save is an Openclaw Skills memory system that turns live conversations into durable, searchable local knowledge.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install save
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install save using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Save — Conversation Memory Index?
Save is an Openclaw Skills workflow for converting ephemeral agent chats into structured memory: a standalone markdown summary, a human-readable conversation log, and a fast local FTS5 search index. It runs entirely on local files and Python stdlib, with zero API keys, zero cloud services, and no pip dependencies.
Designed for Openclaw Skills agents that need persistence without external infrastructure, Save creates a three-tier memory layer: a compact log for immediate context, a searchable index for retrieval, and the full saved file for deep review. The result is a practical knowledge base that makes past decisions, debugging notes, and project context easy to recover in later sessions.
Save — Conversation Memory Index Use Cases
- Preserve important decisions, action items, and session outcomes after a chat ends.
- Save research findings, debugging conclusions, or architecture notes as reusable knowledge.
- Rebuild context quickly before starting a new task by searching prior conversations.
- Maintain a local, self-contained memory layer for Openclaw Skills without external services.
- Handle special cases like custom filenames, long sessions, or off-the-record saves.
- Keep a synced conversation log that shows what has been stored at a glance.
How Save — Conversation Memory Index Works
- The agent detects a save request such as "save this conversation" or
/save. - It synthesizes a clean, standalone summary from the current session context, stripping tool noise, system messages, and verbatim chat logs.
- It writes the result to
~/.openclaw/workspace/saved/YYYY-MM-DD_topic-slug.md. - It prepends a new cross-reference entry to
conversation-log.mdwith the date, filename, and one-line summary, then updates the conversation count. - It rebuilds the local FTS5 index with
memory-index.py build, which scans tracked markdown files, generates the search database, and refreshes the topic map and stub context files. - When users later search or inspect memory, the agent uses the log for quick recall, the index for fast lookup, and the full file for complete context.
Save — Conversation Memory Index Setup
Prerequisites:
- Python 3.8+ with stdlib only
- An OpenClaw agent with read, write, edit, and exec tools
Initial setup:
mkdir -p ~/.openclaw/workspace/saved
Initialize the conversation log:
echo -e '# Conversation Log\n_0 conversations_\n' > ~/.openclaw/workspace/saved/conversation-log.md
Build the initial search index:
python3 {baseDir}/memory-index.py build
Optional environment overrides:
OPENCLAW_WORKSPACEdefaults to~/.openclaw/workspaceSAVED_CONVERSATIONS_DIRdefaults to~/.openclaw/workspace/saved
Manual verification and maintenance:
python3 ~/.openclaw/workspace/saved/memory-index.py search "your query here"
python3 ~/.openclaw/workspace/saved/memory-index.py tags
In normal use, the Openclaw Skills agent handles the save flow automatically: write the file, update the log, and rebuild the index.
Save — Conversation Memory Index Data Schema & Taxonomy
Core storage layout
| Artifact | Location | Purpose | Notes |
|---|---|---|---|
| Saved conversation files | ~/.openclaw/workspace/saved/YYYY-MM-DD_topic-slug.md |
Durable session summaries | One file per save, named by date plus kebab-case topic slug |
| Conversation log | ~/.openclaw/workspace/saved/conversation-log.md |
Human-readable cross-reference index | Date-sorted, prepended on each save, keeps a running conversation count |
| FTS5 search database | /dev/shm/memory-index.db |
Fast local full-text search | Built on tmpfs for speed; uses BM25 ranking and stemming |
| Topic map JSON | Generated by memory-index.py |
Maps topics to file paths and snippets | Used for retrieval and quick context assembly |
| Stub index markdown | Generated by memory-index.py |
Compact inline context for agents | Small enough to fit comfortably in prompt context |
Saved file structure
- Title and topic summary
- Concise context recap
- Decisions made
- Action items and follow-ups
- Relevant notes, findings, or links
- Optional custom filename when the user specifies one
Metadata taxonomy
- Date prefix:
YYYY-MM-DDfor chronological sorting - Topic slug: lowercase kebab-case descriptor derived from the session subject
- Entry summary: one-line description in
conversation-log.md - Search metadata: topic tags, synonym-expanded terms, and snippets generated by the indexer
- Status handling: normal save, off-the-record save, duplicate-suffixed save, or long-conversation summary
Consistency rules
- The log must stay in sync with files on disk
- Duplicate filenames get a numeric suffix such as
-2 - Long conversations can be summarized instead of fully copied
- Reconcile tooling checks for orphan entries and missing files
Save — Conversation Memory Index Advanced Features
- Three-tier memory architecture: always-on conversation log, on-demand FTS5 index, and full-file retrieval.
- Fast local search with BM25 ranking, Porter stemming, and Unicode support via
unicode61. - Synonym-aware query expansion across multiple concept groups to improve recall in Openclaw Skills workflows.
- Topic map JSON generation for topic-to-file lookup and snippet-based retrieval.
- Stub index markdown generation so the agent can keep a compact memory summary in context.
- Fallback grep search when FTS5 returns no matches, improving resilience for edge queries.
- Off-the-record mode: save the file but skip log updates and index rebuilds when requested.
- Duplicate filename protection with automatic counter suffixes.
- Long-conversation handling that switches to executive summaries when the source material is too large.
- Local-only operation with no external dependencies, no API keys, and no cloud services.
- Rebuild-on-save automation that keeps Openclaw Skills memory fresh after each session.
- Reconciliation-friendly design that helps validate log/file consistency and prevent orphaned references.
SKILL.md
Loading
Related Openclaw Skills
savemaintenance
savemaintenance is an Openclaw Skills maintenance pipeline that backs up, reconciles, rebuilds, and audits the three-tier memory system in one deterministic run.
Dali Creative Intelligence
Dali scores and upgrades image and video prompts before generation, preventing wasted credits and improving output quality.
OpenClaw VPS Security Preflight
A read-only VPS security preflight that validates OpenClaw exposure, SSH/firewall posture, and deployment readiness on authorized Linux hosts.
Last Mover Advantage
A strategy skill that tests whether a market rewards being first, or rewards the durable last mover that wins the endgame.
GIF Picker
GIF Picker lets Openclaw agents choose a local reaction GIF by tag and return a reliable `MEDIA:` path for chat replies.
UI Master Static
Openclaw Skills guidance for building production-grade interfaces with native CSS, BEM, modern layouts, and polished motion.
Featured*
