| title | Glossary | |||
|---|---|---|---|---|
| nav_order | 5 | |||
| tags |
|
|||
| aliases |
|
|||
| status | current |
This glossary defines terms, acronyms, and concepts used throughout AutoBot documentation and codebase. For how the platform-level terms fit together, see The AutoBot Platform Model.
A document that captures an important architectural decision along with its context and consequences. See docs/adr/.
An autonomous AI component that can perform specific tasks. AutoBot uses multiple specialized agents including KB Librarian, RAG Agent, and Task Agents.
The central entry point for all API requests, handling routing, authentication, and rate limiting. Located at <backend-ip>:8443 (HTTPS).
A non-blocking Redis or HTTP client that allows concurrent operations. Used for high-performance data access.
The FastAPI-based REST API service running on the main machine. Handles all business logic, LLM orchestration, and data management.
The browser role (<browser-ip>:3000) dedicated to Playwright browser automation. Isolated for security and stability.
The message processing pipeline from user input to AI response. Includes intent routing, context retrieval, LLM inference, and streaming output.
Vector database used for code analysis and duplicate detection. Separate from the main knowledge base which uses Redis.
A design pattern that prevents cascading failures by stopping requests to failing services. Implemented in src/circuit_breaker.py.
The maximum amount of text an LLM can process in a single request. Varies by model (4K to 200K tokens).
AutoBot's distributed, role-based infrastructure design where each role serves a specific purpose — Docker, a single VM, or however many machines a deployment scales to. See ADR-001.
Procedures and documentation for recovering from system failures. Critical for production deployments.
A numerical vector representation of text or code. Used for similarity search, RAG, and duplicate detection.
Advanced search functionality with filtering, reranking, clustering, and query expansion. See src/knowledge/search.py.
The Python web framework used for the backend API. Provides async support, automatic OpenAPI docs, and type validation.
A code smell where a method accesses data from another object more than its own. Indicates misplaced logic.
The set of nodes (machines or containers) that AutoBot's Service Lifecycle Manager (SLM) deploys, operates, and scales — vector DB, cache, database, inference, and workers across one or more hosts.
The frontend role (<frontend-ip>:5173) - the only place allowed to run the Vite frontend server.
A code smell where a class has too many responsibilities. Should be refactored into smaller, focused classes.
The platform-core controls that every layer above inherits: role-based access control (RBAC), review/approval gates, and budgets. Modules such as AutoBot LLC rely on this rather than implementing their own.
The ability to continue operating at reduced functionality when components fail.
Vite's ability to update modules in the browser without full page reload during development.
The platform's extension points — places where modules and plugins attach behavior to core events and workflows. Part of how a module is built on the platform's bones.
Adding more instances of a service to handle increased load. Contrast with vertical scaling.
The platform's persistent, shared knowledge: the RAG knowledge base plus the memory/knowledge graph. It survives across sessions and is shared by everything running on the core, so modules and agents start with organizational context instead of a blank slate.
Neural Processing Unit - dedicated AI acceleration hardware in Intel CPUs. Used for local embedding generation.
The vector-indexed document store containing project documentation, facts, and learned information.
The agent responsible for maintaining the knowledge base, indexing documents, and managing facts.
The RAG (Retrieval-Augmented Generation) framework used for knowledge base queries and document indexing.
AutoBot's flagship module: an autonomous agent-company you install on the platform. It adds companies, org charts, goals, backlogs, sprints, heartbeat scheduling, and board governance, while its agents inherit institutional memory, local inference, hooks, and governance from the core. Here "LLC" names the module — not a legal "limited liability company." See AutoBot LLC.
AI models that generate text responses (run locally or via a provider). AutoBot routes all of them through one LLM gateway with provider fallback. Note: an LLM is not the SLM — see Service Lifecycle Manager (SLM).
Running model inference on hardware you own (CPU, GPU, or NPU) instead of a remote API. A platform-core capability that gives modules and agents inference at zero marginal cost per request.
Tasks that exceed normal request timeouts. Managed by the async task framework with progress tracking.
The system for storing and retrieving context, decisions, and findings across sessions.
An installable capability built on the AutoBot platform core. A module inherits the core's primitives — institutional memory, local inference, hooks, and governance — so it stays small relative to what it delivers. AutoBot LLC is the flagship module. See The AutoBot Platform Model.
AI capabilities that span multiple modalities: text, image, voice, and desktop interaction.
Redis database accessed by logical name (e.g., "knowledge") rather than number (e.g., db=1). See ADR-002.
The NPU worker role (<npu-ip>:8081) - dedicated service for Intel NPU-accelerated AI inference.
Local LLM inference server running on the AI/ML role. Provides self-hosted model inference.
Intel's toolkit for optimizing AI models for Intel hardware including NPUs.
AutoBot's small, stable core — chat/streaming, the knowledge base (RAG + memory graph), the LLM gateway, local inference, hooks, and governance. It changes slowly so the SLM and modules can depend on it. Embodies the platform's promise: your data stays on your machines and the AI stays yours.
Browser automation framework running on the browser role. Used for web scraping and UI testing.
Scripts that run before git commits to enforce code quality, formatting, and security checks.
AI technique that retrieves relevant context before generating responses, reducing hallucinations.
Enhanced Redis with modules for vectors, JSON, and search. Running on the database role.
Post-processing search results to improve relevance ranking using AI models.
Maximum acceptable time to restore service after failure.
Maximum acceptable data loss measured in time.
AutoBot's management layer. It deploys, operates, and scales the infrastructure behind private AI — vector DB, cache, database, inference server, and workers — across a fleet: deploy (stand up the stack, via Ansible) → operate (upgrade, monitor, recover, rotate certs) → scale (add nodes, add NPU workers, assign roles). SLM always means Service Lifecycle Manager — never "small language model" (for on-device models, see Local Inference and LLM). See The AutoBot Platform Model.
A user's conversation context including message history and state. Persisted in Redis.
Real-time token-by-token output from LLMs via WebSocket.
Utilities (sync-to-vm.sh) that synchronize code from local development to VMs.
Dataclass pattern for bundling multiple parameters into a single context object. Reduces long parameter lists.
Claude Code's task tracking tool for managing immediate work items.
Character encoding standard. All AutoBot file I/O must use explicit UTF-8 encoding.
Database optimized for storing and searching embedding vectors. Redis (knowledge base) and ChromaDB (code analysis).
Increasing resources (CPU, RAM) on existing machines. Contrast with horizontal scaling.
Frontend build tool and development server. Only runs on the frontend role.
Virtual Network Computing - desktop streaming protocol. AutoBot exposes VNC at port 6080.
Isolated computing environments. AutoBot is hypervisor-agnostic and can deploy onto one or more hosts or VMs (co-located or distributed); it requires only a supported OS and hardware.
Frontend JavaScript framework (Vue 3) used for the AutoBot web interface.
Bidirectional communication protocol used for chat streaming and real-time updates.
Automated multi-step task execution. Managed by the Enhanced Orchestrator.
Windows Subsystem for Linux - where the main AutoBot backend runs.
| Acronym | Full Form |
|---|---|
| ADR | Architecture Decision Record |
| API | Application Programming Interface |
| DR | Disaster Recovery |
| HMR | Hot Module Replacement |
| KB | Knowledge Base |
| LLC | AutoBot LLC (autonomous agent-company module) |
| LLM | Large Language Model |
| MCP | Memory Context Protocol |
| NPU | Neural Processing Unit |
| RAG | Retrieval-Augmented Generation |
| RTO | Recovery Time Objective |
| RPO | Recovery Point Objective |
| SLM | Service Lifecycle Manager (not small language model) |
| VM | Virtual Machine |
| VNC | Virtual Network Computing |
| WSL | Windows Subsystem for Linux |
Author: mrveiss Copyright: © 2025 mrveiss