Chat left sidebar — Skills, Prompts, Memory, Bookmarks

URL: https://chat.omnilore.ai
Audience: signed-in members
Scope: LibreChat UI features in/near the left sidebar and related panels.
Not: OmniLore fabric memory (SoT / tribal / Goldie) — that is separate (see §5).

LibreChat UI labels can move slightly with version updates. If a name differs, use the closest match.

1. What is dogfooded vs what is convenience UI

| Feature | Enabled on this install? | Member how-to (this doc) | Full product dogfood? |
|---------|--------------------------|---------------------------|------------------------|
| Modes / Agents / New chat | Yes | USER_GUIDE_CHAT_OMNILORE_AI.md | Yes (primary) |
| Bookmarks | On (LibreChat default) | Yes (§2) | Light / operator checklist |
| Prompts | prompts: true | Yes (§3) | Light / operator checklist |
| Skills | Agent capability on | Yes (§4) — with honesty residual | Partial (config + seed; UI residual) |
| Memory | Agent capability on | Yes (§5) — with honesty residual | Partial (config; UI residual) |

Primary product path remains Modes: OmniLore for product truth and tools; Quick for tool-free speed; CodeHub for private git. Sidebar libraries do not replace OmniLore mode for “What is OmniLore?”

2. Bookmarks

What they do

Bookmarks let you star or tag conversations so you can find them again in the sidebar.
They are your account’s conversation labels in Chat (Mongo), not shared org knowledge.

How to use

1. Sign in to https://chat.omnilore.ai
2. In the left conversation list, open a thread (or hover it).
3. Open the conversation ⋯ menu.
4. Choose Bookmark / Add bookmark (wording may vary).
5. Optionally add a tag/name, then save.
6. Open the Bookmarks entry in the sidebar (or filter) to jump back.

When to use

| Use | Skip |
|-----|------|
| Long-running project threads | Treating bookmarks as “official product docs” |
| Threads you reopen often | Sharing bookmarks across other users (not dual-seat tribal) |

Limits

• Per your Chat login only.
• Not dual-seat SoT; not Goldie tribal:shared:.
• Deleting a conversation may remove or orphan its bookmark.

3. Prompts (prompt library)

What they do

Prompts are reusable message templates (optionally with fill-in variables).
They paste or submit a canned instruction so you do not retype long preambles.

Config on this fabric: interface.prompts: true in librechat.yaml.

How to use

1. Open the Prompts panel (sidebar icon or side panel — look for “Prompts” / library).
2. New prompt / create → write the text you want reused.
3. Save.
4. Later: click the prompt name to load it into the chat box.
5. If the template has placeholders, fill them, then send.

When to use

| Use | Do not use for |
|-----|---------------------|
| Recurring tone/style instructions | Product truth about OmniLore (use OmniLore mode + tools) |
| Checklists you run often | Replacing agent instructions for forge/MCP work |
| Personal drafting starters | Storing secrets (passwords, tokens) in prompt text |

Limits

• A prompt is not an agent and does not grant tools.
• On Quick mode, prompts still do not add fabric MCP tools.
• Org presets are off on this install so Modes stay clear — that is intentional and separate from your personal prompt library.

4. Skills

Two different “skills” (do not mix them up)

| Kind | Where | What it is |
|------|--------|------------|
| LibreChat Skills (this sidebar) | Chat UI Skills | Reusable SKILL.md-style instruction packs for Agents |
| OpenClaw / operator skills | Operator packs, OpenClaw | Dual-seat ops skills (challenge dual, image-edit, …) — not this panel |

This section is only about LibreChat Skills.

What LibreChat Skills do

A skill is a small instruction bundle an agent can load:

1. Manual — you pick it (often via $ in the message box, if the UI supports it).
2. Model-invoked — the agent selects from a skill catalog when relevant.
3. Always apply — skill content is applied every turn when configured that way.

How to use (member)

1. Prefer OmniLore mode (or an agent that has tools you need).
2. Open Skills from the side panel / agents area.
3. Create or upload a skill (markdown / pack) if your account allows, or use skills already attached to org agents.
4. Invoke manually if offered ($ skill picker), or ask the agent to apply a named skill.
5. Keep skill text short — huge skills worsen context limits (empty_messages risk).

What this fabric actually seeds today

| Agent | Skills field (seed) | Notes |
|-------|---------------------|--------|
| Shared Operator (OmniLore mode) | omnilore-frontier (+ skills_enabled: true) | Name may appear in agent config; UI catalog load can be residual |
| Ops / Public Law agents | often empty / disabled | Prefer their instructions + MCP tools |

Honesty residual: Skill names in agent JSON are not always fully loaded as first-class LibreChat Skill catalog entries. For reliable product work, trust Modes + MCP tools + agent instructions, not skill names alone.

When to use

| Use | Prefer instead |
|-----|----------------|
| Personal reusable agent playbooks | OmniLore mode for product facts |
| Narrow task recipes on your agent | Full kitchen-sink tool lists (breaks chats) |

5. Memory

Two memory planes (critical)

| Plane | What it stores | Product truth? |
|-------|----------------|----------------|
| LibreChat Memory (sidebar / Memory UI) | Personal notes about you (prefs, ongoing tasks) for chat personalization | No |
| Session feed (fabric) | Searchable summaries of your Chat sessions (session_feed_search) | No — continuity only, not product encyclopedia |
| User posture | Rolling focus from intents/handoff/feed | No — retrieval, not ML |
| OmniLore fabric memory | Disk SoT, tribal/Goldie pointers, optional session-summary search | Yes for product/ops (via tools) |

LibreChat Memory — how it works

When Memory is available in the UI:

1. Open Memory (sidebar or settings / personalize area).
2. Review entries the system inferred, or add / edit / delete manually.
3. Optionally toggle memory for a conversation if a personalize control is shown.
4. Memory may update in the background while you chat (LibreChat design) — still only your personal plane.

What members should not expect

• Memory is not the comprehensive knowledge base.
• Memory is not dual-seat shared tribal.
• Empty vector search ≠ “OmniLore has no docs” — use OmniLore mode and product answer paths.
• Do not put secrets, passwords, or PHI into Memory entries.

Related OmniLore feature (different)

FEATURES_USER.md lists session memory (yours) — optional search of your prior session summaries. That is still personal, not org encyclopedia.

For product definition: TRUTH_AND_GOVERNANCE.md · OmniLore mode.

6. Quick map — which control for which goal

| You want… | Use |
|-----------|-----|
| “What is OmniLore / dual-seat / HOOTL?” | OmniLore mode (not Memory, not a Prompt alone) |
| Fast code snippet, no tools | Quick mode |
| Private git write | CodeHub |
| Reuse a long instruction text | Prompts |
| Star a thread for later | Bookmarks |
| Personal prefs the bot should recall | Memory (personal only) |
| Agent instruction packs | Skills (with residual honesty above) |
| Private specialist agent | AGENT_BUILDER.md / My Agents (small tool list) |

7. Privacy and support

• Bookmarks, Prompts, Memory, and Skills content live with the Chat product for your login.
• Shared fabric knowledge uses rights-based membership and SoT — not these panels.
• Problems: SUPPORT.md · KNOWN_LIMITATIONS.md

End.

Related (fabric continuity, not sidebar Memory)

• Ops SoT map: docs/SESSION_FEED_AND_AGENT_MEMORY_PLANES.md
• Member: PROJECT_INTENT_POSTURE_AND_CONTINUITY.md