ComradeIQ is an AI mission control. You give Commander Atlas an objective, and a coordinated team of specialists (Researcher, Writer, Critic, Formatter, Assembler) works through a real dependency pipeline to hand you back a downloadable result, while you watch every step happen live.
Most "multi-agent" AI demos are theater: spinning cards that invent their own progress, phantom parallel calls, and confident answers with no traceable source. ComradeIQ takes the opposite stance. Every visible update traces to a real event, every artifact to real storage, and every error is an honest error. When a capability is not configured, the product says so plainly instead of faking it.
Watch the whole team work at once: a live agent graph lights up as the Commander delegates, a terminal-style agent console streams the real orchestration events, and the answer arrives with copy, share, and download actions.
Give one objective and ComradeIQ routes it to the right workflow automatically:
| Mode | You get |
|---|---|
| Direct chat | A fast, single answer with no unnecessary overhead |
| Document | A downloadable Markdown file (README, spec, report, docs) |
| Presentation | A downloadable PPTX deck in one of 4 themes (Camo, Cyberpunk, Minimal, Ocean) |
| Research | A sourced, cited answer with real links (opt-in web access) |
Every session starts from a calm command center with one-click example missions across all four modes:
For work that needs coordination, the Commander builds and executes a real dependency DAG, where each specialist waits for its actual upstream output and sees only its own inputs (no shared transcript):
Presentation missions run the full pipeline and generate a real, downloadable PPTX with a live in-app slide preview.
 |
 |
Say "play chess with me" and a rules-validated board opens in chat. You play White; Commander Atlas plays Black through the same model that runs missions.
Ask for a video and ComradeIQ searches YouTube and embeds a playable player right in the conversation.
 |
 |
The Commander connects directly to each specialist (never specialist-to-specialist), and every mission has a full, honest activity view.
 |
 |
Finished missions also get a shareable read-only /m/<id> permalink, and every conversation has a hover menu to share, archive, or delete it.
- Next.js 15, React 19, TypeScript, Tailwind CSS for a responsive, type-safe chat workspace
- Any OpenAI-compatible provider for server-only model calls; the live deployment runs Groq
llama-3.3-70b - Server-Sent Events for live progress, with optional Ably as a progressive enhancement
- Vercel Blob private storage for durable mission records and artifact bytes
- PptxGenJS for server-side decks, chess.js for a legality-validated board with an LLM opponent, and a keyless YouTube search + embed for in-chat video
- Zustand for client state, Vitest + Playwright for unit, DAG, state, accessibility, and reconnect coverage
| Route | Purpose |
|---|---|
/ |
Marketing landing page (shown first) |
/app |
The ComradeIQ mission-control tool |
/m/[missionId] |
Public, read-only shared mission result |
Prerequisites: Node.js 22.13+ and npm, plus an API key for any OpenAI-compatible provider (a Groq gsk_... key works out of the box).
git clone https://github.com/ShivankXD/ComradeIQ.git
cd ComradeIQ
npm ci
cp .env.example .env.local # then add your key
npm run devOpen http://localhost:3000. Without a key, ComradeIQ shows an honest setup state instead of faking a response.
A single key is enough to run live. A gsk_... key is auto-detected and configured for Groq:
OPENAI_API_KEY=gsk_your_key_hereFull options are documented in .env.example. Deployment guidance (including the Vercel Blob store that makes replies and downloads durable across serverless instances) is in VERCEL_SETUP.md.
app/ Landing page, /app tool, /m share page, API routes
components/ Chat workspace, agent graph, agent console, chess, video, panels
lib/ Orchestrator, DAG, provider client, intent routing, storage, store
docs/ Architecture, testing, screenshots, gallery
tests/ Vitest unit + Playwright e2e
npm run lint
npm run test
npm run buildBrowser coverage:
npx playwright install chromium
npm run test:e2e- Provider, Ably, and Blob credentials are server-only and never exposed to the browser.
- An HTTP-only anonymous session cookie owns missions, events, retries, cancellation, and artifacts.
- State-changing requests require same-origin checks; artifact access is validated against mission ownership.
- The API applies input validation, MIME and size limits, rate and concurrency limits, timeouts, cancellation, and basic moderation.
Released under the MIT License.