Norman Agent
Overview

Norman Agent

AI agent runtime with tools, memory, and task execution. Norman Agent adds intelligence on top of Norman Engine by providing conversational memory, tool integration, and multi-step task decomposition.

What It Does

  • Conversational Memory — Short-term (conversation history) and long-term (persistent knowledge) memory powered by MongoDB
  • Tool System — Pluggable tool registry with schema-based tool definitions, auto-discovery, and execution
  • Task Agent (v1) — Complex task decomposition into parallel steps with dependency management and worker agents
  • Norman Engine Integration — Routes all LLM calls through Norman Engine for unified token tracking

Agent Modes

v0.1 — Simple Chat Agent

Standard chat enhanced with memory and tools. Conversations flow through:

  1. Load conversation history + relevant long-term memories
  2. Inject available tool schemas into system prompt
  3. Forward to Norman Engine for LLM completion
  4. Parse TOOL_CALL:name({params}) patterns from response
  5. Execute tools and integrate results
  6. Store conversation turn in memory

v1 — Parallel Task Agent

Complex task execution with decomposition and parallel workers:

  1. User submits a task with steps and dependencies
  2. Agent creates a dependency graph
  3. Independent steps execute in parallel
  4. Results cascade to dependent steps
  5. Final synthesis produces the deliverable

API Surface

EndpointMethodDescription
/api/chatPOSTChat with memory and tools (streaming + non-streaming)
/api/chat/:chatIdGETRetrieve conversation history
/api/memory/:userIdGETList user memories
/api/memoryPOSTStore a memory
/api/memory/:userId/searchPOSTSearch memories
/api/memory/:userId/summaryGETSummarize memories
/api/tasksGET/POSTList or create tasks (v1 mode)
/api/tasks/:taskIdGETTask status and details
/api/tasks/:taskId/executePOSTStart task execution
/api/toolsGETList available tools
/api/tools/executePOSTExecute a tool directly

© 2026 Shell Technology. All rights reserved.