Jarvis Agent Assistant

Give Copilot access to your work items, pipelines, and designs — without leaving VS Code.

You use it by just using Copilot. There's no separate chat agent to address and no new syntax to learn. Jarvis plugs into normal Copilot Chat as a set of tools, so when your question needs something that isn't in your code, Copilot reaches through Jarvis to get it — instead of guessing from the files in front of it. Nothing changes until you say so.

You, in normal Copilot Chat
  │   "why is the build red?"   "create a story for this"   "what's assigned to me?"
  ▼
GitHub Copilot ── realises the answer isn't in the workspace, calls a Jarvis tool
  ▼
Jarvis ── your identity, your sign-ins, nothing hosted in between
  │
  ├─► Azure DevOps — work items, pipelines, pull requests
  ├─► Jira — issues, status transitions, comments
  ├─► GitHub — issues and pull requests
  └─► Figma — frames, layers, and text from a design link
  ▼
Copilot answers with the real facts — and asks you to confirm before it changes anything

Ask "what am I working on?", "why did CI fail?", "create a user story for this bug", "implement 4011482", or paste a Figma link and say "build this screen". Copilot handles the conversation and writes the code; Jarvis reaches your tools. Jarvis also hands Copilot ready-authenticated Azure DevOps and GitHub MCP servers, so agent mode gets those too — with no mcp.json to edit and nothing extra to install.

Use agent mode. Jarvis contributes language-model tools, and Copilot only offers tools to the agent — in ask mode they aren't in its picker.

What it does

• Azure DevOps, Jira & GitHub — name a work item id (4011482), a Jira issue (PROJ-123), or a GitHub issue (owner/repo#123) and Copilot fetches the real description, acceptance criteria, and discussion before it writes a line of code.
• Create and update work — ask Copilot to raise a task, user story, bug, or epic in Azure Boards or Jira, break work into sub-tasks, move something to Done, or comment on a ticket. Jarvis reads the project's own required fields first, so items land valid the first time — and it remembers any rejection so the next attempt already knows.
• Pipelines & CI — ask why a build is red and get the failed steps, log tails, and the commits in the run, not a guess from the YAML.
• Your day at a glance — a sidebar dashboard with a list per platform: Azure DevOps work items, Jira issues, Azure Repos pull requests, GitHub pull requests. Click a row to open it in Copilot Chat, or press the calendar button for a spoken-word briefing.
• Background mentions — comment @jarvis on your Azure DevOps work items or Jira issues and Jarvis notices, then offers to open that task in Copilot Chat so you can pick it up from VS Code.
• Figma context — paste a Figma link alongside a task to fold the real frames, layers, and text into what Copilot sees.
• Pull requests, properly — ask Copilot to open the PR and Jarvis does it against your actual origin, using your VS Code GitHub sign-in. No gh in a terminal, no invented owner or repo.
• Learns from failures — when something goes wrong (a rejected push, a missing required field), Jarvis records the fix in your workspace and applies it next time instead of repeating the mistake.
• Stops Copilot re-reading your repo — instead of crawling the tree on every task, Copilot asks Jarvis for a repository map: structure, stack, entry points, where the tests live, and the files past work actually touched. One call replaces the whole "figure out where I am" step.
• Remembers solved tasks — when you open a pull request, Jarvis offers to remember the task: what it was, how you solved it, and which files you touched. Next time something similar comes up, Copilot is handed that summary and file list up front, so it reads a few known files instead of searching the whole repository. Off by default; you approve every save.

Why you'll like it

• Nothing to learn — it's the Copilot Chat you already use. Jarvis just widens what it can see.
• You stay in control — every read is free and every write asks first, naming exactly what it will do.
• Grounded in your code and your tickets — answers come from your actual repository and the real work item, not from a plausible guess.
• Uses your own Copilot — runs on your identity and your GitHub Copilot seat, with whichever model you picked in the chat model picker. No hosted services, no shared storage.
• It remembers, so it gets cheaper — solve a task, save it, and the next similar one starts from your solution and the files you edited instead of exploring the repository again. Opt-in, on-device, and nothing ever leaves your machine.

Getting started

1. Install the extension and reload VS Code.
2. Open the repository you want to work in.
3. Configure your task tracker:
   • Azure DevOps — you'll be prompted for your organization and project on first run.
   • Jira — run Jarvis: Connect Jira and paste your site URL (or any ticket link on it).
   • GitHub — nothing to do; it reuses your VS Code GitHub sign-in.
4. Open Copilot Chat, switch to agent mode, and ask:
   • "What am I working on?"
   • "Why did the last build fail?"
   • "Create a user story for adding a logout button."
   • "Implement 4011482" — or any Jira key, GitHub issue, or Figma link.

Requirements

• VS Code 1.101 or later
• An active GitHub Copilot subscription
• An Azure DevOps, Jira, and/or GitHub account with tasks assigned to you

Privacy & security

Jarvis acts as you, using your own identity and Copilot seat. Nothing is hosted, shared, or stored outside your machine and the services you already use.

Nothing changes without your click. Reading is free — work items, issues, pipelines, designs. Every write is gated:

• Creating a work item, updating one, opening a pull request, or saving a lesson or task memory shows a confirmation naming exactly what will happen. Decline and nothing occurs.
• Jarvis never writes to your board or your remote on its own initiative, and the Azure DevOps connector it shares with Copilot is read-only unless you opt in.

The optional memory layer is local-only and redacts secrets before storing anything.

Full technical documentation, architecture, and configuration reference are available in the project repository.