Agent Workspace

A no-code platform where users create their own AI agents, connect multiple MCP integrations and knowledge bases, and direct them to perform multi-step tasks - only when asked.

Comment

Agent Workspace

Build, connect, and deploy custom AI agents — on your terms.

One-Line Pitch

A no-code platform where users create their own AI agents, connect multiple MCP integrations and knowledge bases, and direct them to perform multi-step tasks — only when asked.

1. The Problem

Every team needs an assistant that can actually do work across the tools they already use — not just answer questions.

Today, employees face three bad options:

  1. Generic chatbots — can summarize text but cannot call live APIs, read from Slack, update a wiki, or run a scheduled job.
  2. Custom scripts — fragile, engineer-maintained, and break when tools or credentials change.
  3. Point solutions — one bot for support, another for docs, none configurable by the person who knows the workflow.

The real need is personal and team-level: a user should be able to create an agent, give it access to the MCPs and knowledge sources it needs, and then ask it to perform tasks — compare data across systems, generate reports, update records, set up a recurring job — without writing code.

Agents must be user-directed. Nothing runs in the background unless the user explicitly requests it, including any schedule they define in conversation.

2. Our Solution and How It Works

What we are building

Agent Workspace is an open-source platform where any user can:

Step What the user does What the platform does
Configure Create an agent in a UI — write instructions, upload static files (PDF, MD, TXT), connect one or more MCP integrations, choose allowed tools Stores agent config, indexes knowledge, registers MCP connections
Direct Chat in plain language: "Pull last week's data from X, compare with Y, and give me a report" Orchestrator plans steps, retrieves from knowledge base, calls MCP tools, returns results with full audit trace
Schedule (optional) Ask: "Run this every Monday at 9am" Creates a user-defined cron job — only because the user asked
Deploy Enable Slack (or other channel) for the same agent Same runtime, same config — web chat or @mention in Slack

One agent. Multiple MCPs. Static knowledge plus live tool access. Retrieval plus action. All triggered by the user.

Core capabilities

  • Multi-MCP connectivity — plug in Slack, Confluence, GitHub, filesystem, and any MCP-compatible server; mix and match per agent.
  • Knowledge base (RAG) — upload documents so the agent grounds answers and decisions in your content, not just live API data.
  • Tool calling + retrieval — the agent decides when to search docs vs call a live tool, in a multi-step loop.
  • User-initiated execution only — chat commands, manual "run now," or schedules the user explicitly creates; no silent background automation.
  • Deploy anywhere — start in the web UI; deploy the same agent as a Slack bot or other integrated channel.
  • Full audit trail — every tool call, retrieval, and output logged so users trust what the agent did.

How it works

User (Web UI or Slack)
        ↓
   Agent Orchestrator
        ↓
   ┌────┴─────┬────────────┬──────────────┐
   LLM      Knowledge     MCP Tools      Scheduler
            Base (RAG)    (any connected)  (user-defined)

The user owns the agent. The agent executes what it is told — once or on a schedule the user defined.

Example tasks users can ask (not the product — illustrations only)

  • "Compare support questions in our chat channel against our FAQ doc and tell me what's missing."
  • "Every Friday, compile open action items from our project tool and post a summary."
  • "Search our uploaded runbooks and check if this incident matches a known pattern."
  • "Draft an update for our internal wiki based on these meeting notes."

Slack + Confluence FAQ sync is one demo story, not the product scope. The product is the configurable agent layer that makes any such workflow possible without code.

Why this matters for customers

  • Self-serve — the person who knows the workflow configures the agent, not a platform team.
  • Composable — add MCPs and files as needs grow; no rebuild per use case.
  • Trustworthy — user-directed execution and visible audit logs; writes can require confirmation.
  • Deployable — runs locally (Docker + Ollama) for demo and development; open source for teams to extend.

3. Why We Are the Right Team to Build This

We have built and used agent platforms that combine MCP, retrieval, and tool execution in production environments. Our focus for the hackathon: a working, locally runnable prototype that shows agent creation, multi-MCP wiring, knowledge upload, user-directed task execution, and channel deployment - not a single hard-coded workflow.

Built With

Share this project:

Updates