TaskForge: Multi-Agent Productivity Companion

Inspiration

In modern workspaces, we constantly switch between tabs, Slack, emails, and project boards. While single AI chat screens are helpful, they can't handle complex, multi-step workflows on their own. We envisioned "TaskForge"—a smart productivity companion that coordinates multiple dedicated subagents to handle complex daily tasks end-to-end, so you can focus on real high-value work.

What it does

TaskForge is a browser extension and dashboard powered by GPT-5.6 Sol and Codex. It takes a single high-level command (e.g., "Find the latest quarterly marketing draft, extract the action items, and schedule a 30-minute sync with the design lead on my calendar") and instantly delegates it to specialized subagents. A "Researcher Agent" fetches the file, a "Synthesizer Agent" extracts the items, and a "Scheduler Agent" hooks into your calendar to finish the job.

How we built it

The frontend dashboard is built with Next.js and Tailwind CSS to show real-time agent execution trees. The orchestration engine is built on Node.js using LangGraph and the new GPT-5.6 Sol API, which acts as the lead controller. We also utilized OpenAI Codex to dynamically write and execute custom data-parsing scripts whenever our agents need to handle unformatted or legacy documents.

Challenges we ran into

The biggest challenge was preventing "agent loops" and excessive token consumption when agents collaborate. We solved this by implementing a strict execution guardrail system and state management framework. We also designed a human-in-the-loop (HITL) authorization step for high-risk actions like sending emails or booking meetings.

What's next for TaskForge

We want to enable "offline agent tasks" where long-running research workflows can execute on edge servers overnight and deliver a clean summary report to the user's inbox first thing in the morning.

Built With

• gpt-5.6
• langgraph
• next.js
• openai-codex
• tailwind-css