Overview
Contextual AI’s Agent Composer is a framework that enables platform users to build customized query workflows by assembling pre-made components such as search tools, third-party API/MCP integrations, and various LLMs. By combining these primitives into arbitrary computational graphs, users can design specialized agents tailored to their needs.Why Agent Composer?
Engineering teams across semiconductor, manufacturing, energy, and other highly technical industries spend significant time on complex and repetitive work—analyzing device logs, planning production, resolving customer engineering tickets, and more. These tasks demand deep domain expertise, yet occur often enough that the right automation can deliver transformative results. Until now, AI hasn’t delivered on that promise. Generic AI tools struggle with specialized technical content and often hallucinate. Custom-built solutions take months to deploy and quickly break as underlying data evolves. Agent frameworks provide building blocks, but leave teams to stitch together fragile, production-grade systems on their own. What’s been missing is context: a reliable way for AI to access and reason over the technical documentation, specifications, and institutional knowledge engineers rely on every day. Agent Composer is Contextual AI’s answer: the infrastructure and orchestration layer that manages context, enforces guardrails, and maintains accuracy throughout multi-step engineering workflows.What Makes Agent Composer Unique
Unlike simple workflow tools, Agent Composer supports hybrid agentic behavior:- Static Workflows: Defined steps, loops, and branches where you control the recipe
- Agentic Research Steps: LLM-powered research agents that operate within a structured agent loop, using explicitly defined tools and constraints to plan, retrieve information, and determine the best approach for answering a query.
Real World Results
Teams using Agent Composer are seeing dramatic improvements:| Industry | Task | Before | With Agent Composer |
|---|---|---|---|
| Semiconductors | Customer engineering issue resolution | 8 hours | 10 minutes |
| Electronics | Device log analysis | 10 hours | 20 minutes |
| Logistics | Technical docs Q&A | 5 hours | 5 minutes |
| Manufacturing | Production planning | 10 hours | 5 minutes |
Agent Composer Workflows
You can create Agent Composer workflows by building them visually with the GUI or editing the YAML file manually.Agent Composer GUI
The Agent Composer GUI offers a visual workflow builder that lets users design workflows graphically, which are then automatically translated into the corresponding YAML.Agent Composer YAML
Users can create Agent Composer workflows programmatically by providing a YAML file that defines the workflow graph and its configuration. This file outlines the required syntax, framework rules, and expected behaviors. The YAML can be entered directly into the text field under the Advanced Context tab when creating an agent.Tools Included
The following tools can be added to your workflows to help your agents carry out a variety of tasks and functions.- Retrieval tools: Search vector databases, structured data sources, MCP servers, enterprise applications, and any external API read call
- Ingestion tools: Parse, extraction, ingestion, enterprise data connections
- Action tools: Connect to business applications, webhooks
- Planning Tools: Query reformulation, decomposition
- Evaluation Tools: LMUnit, groundedness checks
- Memory Tools: Chat history, multi-turn manager, and more to come