Virtual Employees

---

title: "Virtual Employees" tier: "2" status: "verified" version: "1.2.0" last-verified: "2026-06-17" commit: "b1c347b1" network-badge: "none" risk-tags:

• "RISK: MEDIUM"

---

👥 Virtual Employees

This page covers the creation and configuration of custom virtual employee profiles within Tadpole OS. Virtual employees are AI agents assigned to specific roles, departments, and projects.

---

1. Agent Configuration Panel (AgentConfigPanel)

The primary interface for designing custom virtual employee profiles.

1.1 Cognition Tab

• [Tab] Cognition Tab

  • Default state: Active/focused slot.
  • Visible location: Top-left header of the AgentConfigPanel.
  • Observable side effect: Focuses the active model slot configurations (Primary, Secondary, Tertiary).
  • Keyboard shortcut: None.
  • Missing in build: Always present.

• [Toggle] Long-Term Memory Toggle

  • Default state: Enabled (true).
  • Visible location: Memory section under the Cognition tab.
  • Observable side effect: Enforces saving agent conversation context into LanceDB vector memory.
  • Keyboard shortcut: None.
  • Missing in build: Always present.

• [Input] Working Memory

  • Default state: Ephemeral JSON scratchpad ({}).
  • Visible location: Text area block in the memory panel.
  • Observable side effect: Manually edits the agent's short-term context state.
  • Keyboard shortcut: None.
  • Missing in build: Always present.

1.2 Governance Tab

• [Input] Budget Cap

  • Default state: 1.0 USD (or defined by DEFAULT_AGENT_BUDGET_USD [RISK: MEDIUM]).
  • Visible location: Security governance section.
  • Observable side effect: Restricts the agent's total spending per active mission.
  • Keyboard shortcut: None.
  • Missing in build: Always present.

• [Toggle] Requires Oversight

  • Default state: Enabled (true).
  • Visible location: Security governance section.
  • Observable side effect: Redirects all write-tool executions to the Oversight Queue.
  • Keyboard shortcut: None.
  • Missing in build: Always present.

1.3 Inference Tuning

• [Selector] Model Assignment

  • Default state: First available Ollama model.
  • Visible location: Model slot dropdown selector.
  • Observable side effect: Switches the active LLM provider and model ID for the active slot.
  • Keyboard shortcut: None.
  • Missing in build: Always present.

• [Slider] Temperature

  • Default state: 0.7 (or default of the model).
  • Visible location: Mid panel of the Cognition configurations.
  • Observable side effect: Sets model creativity slider. Range 0.00 to 2.00 with 0.05 steps.
  • Keyboard shortcut: None.
  • Missing in build: Always present.

• [Input] Max Tokens

  • Default state: 4096 (or model default).
  • Visible location: Next to the temperature slider.
  • Observable side effect: Restricts response length per API call.
  • Keyboard shortcut: None.
  • Missing in build: Always present.

• [Input] System Prompt

  • Default state: Standard agent directive template.
  • Visible location: System prompt textarea block.
  • Observable side effect: Defines the role identity instructions for the agent model.
  • Keyboard shortcut: None.
  • Missing in build: Always present.

1.4 Reasoning Engine (Mythos)

Tune the agent's internal monologue and reasoning settings (verified in ModelSlotConfig.tsx:L134-164):

• [Slider] Reasoning Depth

  • Default state: 1 turn.
  • Visible location: Mythos parameters box.
  • Observable side effect: Sets the maximum reasoning loop depth. Range 1 to 16 turns with step 1.
  • Keyboard shortcut: None.
  • Missing in build: Always present.

• [Slider] ACT Halting Threshold

  • Default state: 90% (0.90).
  • Visible location: Mythos parameters box.
  • Observable side effect: Defines the confidence level below which the agent continues reasoning. Range 0% to 100% (0.00 to 1.00) with step 0.01.
  • Keyboard shortcut: None.
  • Missing in build: Always present.

---

2. Agent Authority Hierarchy

Agents operate within a strict authority hierarchy (defined in Glossary#roleauthoritylevel):

Level	Role	Capabilities
Executive	CEO, Overlord	Strategic oversight and delegation across all departments
Management	[[Glossary#department-lead	Department Lead]] (formerly Alpha Node)
Specialist	Standard task executor	Individual tool execution within assigned scope
Observer	Read-only auditor	Monitoring and reporting only, no write operations

---

3. Agent Scope & Sandboxing

Each agent operates within strict boundaries:

• Workspace Root: The workspace_root field in Glossary#runcontext defines the sandbox directory for file operations.
• Filesystem Adapter: A zero-trust adapter (FilesystemAdapter) restricts path traversals to prevent escaping the sandbox.
• Allowed Files: Optional whitelist constraints can limit which specific files an agent can access.
• Safe Mode: When safe_mode is enabled, hazardous commands/actions require explicit user confirmation.

---

4. Skills & Workflows

Agents can be equipped with custom capabilities:

• Skills: Custom script files registered to the agent. Skill manifests are stored in data/skills/*.json and validated by Parity-Guard.
• Workflows: SOP (Standard Operating Procedure) files placed in data/workflows/ that enforce step-by-step execution rules.
• MCP Tools: External tools exposed by active Model Context Protocol servers.

→ See RAG-&-Memory for how SOPs are registered and enforced. → See Approvals-&-Quotas for oversight queue configuration per agent.

---

5. Swarm Depth & Delegation

The Glossary#department-lead can recruit sub-agents to handle specialized subtasks. Delegation is controlled by:

Setting	Default	Description
MAX_SWARM_DEPTH [RISK: MEDIUM]	5	Hard limit on recursive agent-spawning depth.
MAX_AGENTS [RISK: MEDIUM]	50	Registry limit for maximum concurrent agent identities. Code default is 50 (defined in state/mod.rs:L547-551). .env.example ships 100.

Tip

Recommended Max Depth: 3 for standard office hardware. Deeper delegation increases memory usage and latency. The absolute maximum is 5.

---

Complete Lexicon: For the authoritative technical breakdown, see the main repository GLOSSARY.md. Every [[Glossary#term|term]] link on this page resolves to an entry there.