Introduction

Agent Serve is the AI workspace where teams build, deploy, and manage AI agents. Build visually in the workflow builder, conversationally in Chat, or with code through the API. Connect the integration catalog and every major LLM to create agents that automate real work.

How you build

When you open Agent Serve, the first thing you meet is Chat, a conversation over your whole workspace. You describe the system you want, and Agent Serve scaffolds it: workflows, tables, knowledge bases, and the wiring between them. You then open what it built, run it, and refine it.

• Chat. Describe a system in natural language and Agent Serve creates and edits the resources for you. The fastest way to start.
• Workflow builder. Design agent logic visually by connecting blocks in the builder. The clearest way to see and fine-tune what runs.
• API and SDK. Build and trigger agents programmatically.

You can mix these on one system, scaffolding in Chat and then fine-tuning in the builder.

The anatomy of a workspace

An Agent Serve system is not a single chat response. It is a set of resources that live in a workspace and connect to each other. The left rail of the app has entries for Relays, Chat, Spaces, Team spaces, Apps, and Workflows, and the rest of the workspace hangs off them. See The workspace for a tour.

• Relays are AI teammates. You give a Relay a role, the data and tools it may use, and the actions it needs your approval for, then hand it jobs and get finished work back.
• Chat is where you build and manage everything in natural language.
• Spaces is where your team writes docs and draws on whiteboards, with Chat beside you to summarize, draft, and suggest edits.
• Workflows are visual programs made of blocks, where your logic runs. You design them in the workflow builder.
• Agents and tools are how a workflow acts. An agent reasons with a model, and tools let it take actions like sending an email or querying an API.
• Run history, in the Workflows pane, records every workflow run, block by block, so you can verify what happened.
• Team spaces hold your team's shared data and analysis:
  • Data sources connect warehouses, databases, apps, and spreadsheets.
  • Tables hold structured rows your workflows and Relays read, write, and enrich.
  • Library stores the documents and media your team and agents use.
  • Knowledge bases are searchable memory. An agent retrieves relevant passages from your documents to ground its answers.
  • Dashboards turn your data into charts, KPIs, and reports.
  • Workbooks let you query your data sources in SQL or plain English, then save a result as a dataset or a dashboard panel.
  • Metrics, the semantic model, and rules define your numbers and business logic once, so every dashboard and agent uses the same definitions.
  • Monitors watch metrics and rules and alert you when something changes.
• Apps are tools that run inside the workspace, on your own data. Your developers build them with the CLI and SDK. Plugins are apps that also add Chat commands, menu actions, and background work across the workspace.
• Deployments expose a workflow to the outside world as an API, a chat, or an MCP server (Model Context Protocol, a standard way for AI tools to call each other).

What you can build

• AI assistants and chatbots. Conversational agents that search the web, manage calendars, send email, and act on your business systems.
• Business process automation. Data entry, report generation, customer responses, and content workflows.
• Data processing and analysis. Extract from documents, analyze datasets, and sync data across platforms.
• API integration workflows. Unified endpoints that orchestrate multi-service logic and event-driven automation.

Integrations

Agent Serve provides native integrations with supported services:

• AI models. OpenAI, Anthropic, Google Gemini, Groq, Cerebras, and local models via Ollama or VLLM.
• Communication. Gmail, Slack, Microsoft Teams, Telegram, WhatsApp.
• Productivity. Notion, Google Workspace, Airtable.
• Development. GitHub, Jira, Linear, automated browser testing.
• Search and data. Google Search, Perplexity, Firecrawl, Exa AI.
• Databases. PostgreSQL, MySQL, Supabase, Pinecone, Qdrant.

For anything not built in, MCP support connects any external service or tool.

Deployment options

• Cloud-hosted. Launch immediately at agentserve.valnoraelric.com with managed infrastructure, scaling, and observability.
• Self-hosted. Deploy on your own infrastructure with Docker Compose or Kubernetes, with support for local models.

Next steps