jcode in Zed with multiple tabs + parallel agents becomes a research superpower:
🧠 Why This Setup is a Research Cheat Code
1. Parallel Literature & Data Processing
Each Zed tab runs an independent jcode agent. You can have:
- Tab 1 — agent scraping and summarizing papers
- Tab 2 — agent cleaning/analyzing your dataset
- Tab 3 — agent writing up findings or generating visualizations
- Tab 4 — agent cross-referencing citations or running experiments
All simultaneously. No waiting for one task to finish before starting the next.
2. Swarm Intelligence on Your Repo
jcode’s Swarm mode is built for exactly this. Spawn multiple agents in the same research repo and they:
- Automatically coordinate — if Agent A edits a file Agent B has read, B gets notified instantly
- Resolve conflicts natively — no git worktree hacks
- DM each other or broadcast — agents can pass findings between themselves
- One agent becomes the coordinator, others become workers running subtasks in parallel
Think: one agent orchestrates your research pipeline while others execute individual experiments concurrently.
3. Persistent Semantic Memory Across Sessions
Every conversation turn is embedded as a semantic vector. When you come back to a research thread days later, jcode automatically recalls relevant prior context — methodology decisions, dead ends, key findings — without you re-explaining everything. This is huge for long-running research projects.
4. Insane Resource Efficiency = More Parallel Agents
At 10 active sessions, jcode uses ~261 MB RAM. Claude Code uses ~2,300 MB for the same. That means on a normal machine you can run 5–8x more parallel agents with jcode than with alternatives before hitting memory limits.
5. Browser Automation Built In
Need to pull live data, scrape a journal, or interact with a web-based research tool? Each agent tab can run browser automation (via Firefox Agent Bridge) — no separate tooling needed.
6. Side Panel as a Live Research Dashboard
Tell an agent to write findings, diffs, or diagrams directly to the side panel in real time. You get a live-updating view of what the agent is doing — perfect for monitoring long-running analysis without interrupting it.
7. Any Model, Any Provider
Route different agents to different models based on the task:
- Heavy reasoning → Claude or GPT-5.5
- Fast lookups → Gemini Flash or Groq
- Local/private data → Ollama or vLLM on your own machine
All from the same interface, switchable mid-session.
The Research Workflow in Practice
Tab 1 (Coordinator Agent): "Orchestrate a literature review on X"
└── Spawns Tab 2: "Summarize papers 1-20"
└── Spawns Tab 3: "Summarize papers 21-40"
└── Spawns Tab 4: "Find contradictions between summaries"
└── Tab 1: Synthesizes final report as others finish
All agents share memory, coordinate on the same repo, and you watch it unfold in real time across your Zed tabs.
Bottom line: jcode + Zed tabs = a personal research team running 24/7 on your machine, with shared memory, conflict resolution, and the resource footprint of a single lightweight app.