Injected into every agent and CEOcall — keep it factual and concise.
Saved
Previous versions
Proposed updated context
Claude merged your conversation into the existing context, resolving conflicts and removing redundancy. Review before saving.
KB Priority Rules
Applied to all agents — guides how conflicting information across KB files is resolved.
🎯 Operational Modes
Per-thread activity prompts injected into Claude's system prompt. The picker appears when a user creates a new chat thread (the + tab). Edit a mode to refine how it shapes the assistant's behavior — the Prompt fragment field is the actual text Claude reads.
Loading…
Label
Description
Icon
Prompt fragment — injected naked into Claude's system prompt (no header label). The "don't reveal this directive" trailer is appended automatically.
Available for agents — controls which agents show this mode in their new-thread picker.
🗄 Archive prompt
Sent to the active agent when the user clicks Learn from this session in the chat toolbar. The agent uses its workspace_write tools to produce notebook / design notes from the session.
📧 Outlook Accounts
All agents share this allow-list. Only the checked accounts can be read/searched/sent-from. Empty list = NO Outlook access for any agent.
Loading…
▶🔧 Outlook Agent Instructions (Advanced)
System prompt injected into every agent that has Outlook access. Edit this to change how all your agents understand and use Outlook tools. Three placeholders are substituted at runtime:
{{SEND_EMAIL_MODE}},
{{ACCOUNTS_LIST}},
{{TODAY}}.
Changes take effect on the next agent message — no restart needed.
Plain-markdown rules describing folder semantics, indicators, priorities, side-car requirements, and few-shot examples. Injected ONLY into the personal-assistant agent system prompt (other agents ignore this). Read by the agent BEFORE every classification. Length is up to you — Claude handles long rules well; token cost grows linearly. Changes hot-reload on the next agent message.
🗂️ MRP / FileMaker Access
Toggle FileMaker (MRP_VOO) access per agent. Checked = agent gets the FM tools (find, get, create, update, run script, attach image) and the container/link proxies. Unchecked = no FM tools whatsoever for that agent. Changes take effect on the next agent message (no restart needed) — enforcement is server-side at three checkpoints.
Loading…
▶🗂️ MRP Layouts & Field Definitions (what the agent knows)
This text is injected into every FM-enabled agent's system prompt. Describe your layouts, field definitions (with exact case-sensitive names), relations, and query patterns so the agent uses the right fields. It is PURE documentation — it does NOT change server mechanics (the clickable MRP links, container downloads, and write-approval flow keep working regardless). Markdown is fine. Changes take effect on the next agent message — no restart needed.
🌱 Régie de culture — Dimensionnement
Per-agent access to the crop sizing app (multi-sequence surface calculator, 1-4 séquences). Checked = the agent gets the crop_sizing tool (triggered by questions like "donne-moi les surfaces de la régie de culture…") and a "🌱 Régie" tab embedding the interactive app. Unchecked = no tool, no tab. Changes take effect on the next agent message — no restart needed. The app itself stays reachable at /apps/crop-sizing for any signed-in user.
Loading…
🌐 Global Knowledge Sources
Folders watched here feed every agent and VOO. Files are summarized automatically. SHA-256 checksums detect changes.
Add Global Folder
Loading…
🗂 Airtable Scope
Schema, linked-record relationships, and per-table scoping for Airtable data injected into agent prompts.
Loading…
New Airtable Table
Initial fields (the first one is the primary field):
Tables
Click Refresh schema to load.
Relationships
—
Choose an activity for this session
Shapes the assistant's posture for the work ahead. You can change it anytime.
CEO Goals
Enter to create · Esc to cancel
No goals yet. Click + New to create your first CEO goal.
Goal Detail
🧠
Create a goal in the sidebar and click ▶ Run in CEO to deploy agents
Or ask CEO a question below
↵ send · Alt+↵ new line
Select an agent to view and edit its system prompt
Owner-only · Audio and transcripts stay on this device. Run Meetily (Whisper.cpp + Ollama) locally to record and summarise; CEO reads finished summaries straight from Meetily's local database. Each new meeting waits in Pending until you pick a sensitivity, then enters your private team archive.
Meetily database
CEO reads completed summaries from Meetily's local SQLite database. Auto-detected from %APPDATA%. Override only if Meetily is installed elsewhere.
⏸ Pending — sensitivity required
Imported
No meetings imported yet.
Dismissed
⏸ Complete meeting metadata
Leave Title blank to use the auto-suggested phrase (timestamp + topic from the summary). Sensitivity drives who can read the imported entry — Confidential and Internal stay private to you; Shareable is visible to every CEO user.
Preview
📊 User Monitoring
Token intelligence & session analysis
Daily calls
Stacked by user. Click a bar to filter the session list.
