Fixate

Fixate is an AI growth simulator for marketing pages and campaign assets. It lets you test a URL, pasted HTML, or uploaded image before spending on traffic, then returns attention analysis, buyer-response scoring, demographic outreach segments, conversion blockers, improved variants, edited images, and an A/B test plan.

The app uses Meta TRIBE for the heatmap and neural-signal layer, with OpenAI handling the generative and scoring workflow around it:

• Meta TRIBE provides the attention heatmap, fixation regions, and neural-signal indicators used to explain what buyers notice first.
• Meta TRIBE neural signals feed the attention zones, ignored areas, scan-path interpretation, and attention trap analysis.
• OpenAI vision scores buyer-response signals, zones, blockers, and Fixate Score.
• OpenAI agents generate buyer reactions, strategy, creative variants, and A/B plans.
• A Demographics Agent identifies likely outreach segments and tunes the creative path toward the selected audience.
• The same OpenAI model setting (OPENAI_MODEL) drives text, vision, scoring, and image edits through the Responses API image-generation tool.

Local code handles Playwright capture, file storage, SSE streaming, and serving generated artifacts.

For presentations and local development, the backend can expose a Meta TRIBE demo adapter status. This is a no-download compatibility/demo layer for the Meta TRIBE heatmap and neural-signal integration path. It does not load or run a Meta model locally; Fixate keeps using the same OpenAI/local fallback runtime unless a real Meta TRIBE service is connected behind the adapter.

Platform Roles

Fixate also recognizes these external platforms as necessary parts of the broader product architecture and go-to-market workflow:

• Convex provides the realtime backend and database foundation for production-grade collaboration, persistent analysis history, and live workflow state. The frontend now mirrors FastAPI optimization jobs, SSE agent events, scores, heatmap links, and final result metadata into Convex when VITE_CONVEX_URL is configured.
• Orange Slice provides the AI go-to-market and customer-enrichment workflow layer that informs audience discovery, outreach segmentation, and campaign activation.

The current local demo still runs without Convex or Orange Slice credentials. Without a Convex URL, the app stays in local-only mode and skips persistence.

Project Structure

backend/   FastAPI API and optimization pipeline
frontend/  React + Vite UI

Environment

Create backend/.env:

OPENAI_API_KEY=sk-your-key-here
OPENAI_MODEL=gpt-5.4
META_TRIBE_DEMO=false
META_TRIBE_MODEL=meta-tribe-demo-adapter
FIXATE_JOBS_DIR=

FIXATE_JOBS_DIR is optional. Leave it blank to store screenshots, heatmaps, and edited images in your system temp folder.

Set META_TRIBE_DEMO=true when you want the UI and /health endpoint to show the Meta TRIBE heatmap and neural-signal adapter. In the demo adapter mode, the actual inference path remains unchanged.

Create frontend/.env.local after setting up Convex:

VITE_CONVEX_URL=https://your-deployment.convex.cloud

Convex Setup

Convex is used for persisted analysis history and realtime workflow state. FastAPI still runs the optimizer, OpenAI calls, screenshots, heatmaps, image edits, and SSE stream.

cd frontend
npm install
npx convex dev

The first npx convex dev run will ask you to log in, create or choose a Convex project, generate convex/_generated, and print a deployment URL. Put that URL in frontend/.env.local as VITE_CONVEX_URL.

Keep npx convex dev running while developing Convex functions. In a second terminal, run the Vite frontend with npm run dev.

Deploy Convex functions for production:

cd frontend
npx convex deploy

Run Locally

Backend:

cd backend
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
playwright install chromium
uvicorn main:app --reload --port 8080

Frontend:

cd frontend
npm install
npm run dev

Open:

http://127.0.0.1:5173

If you are using Convex persistence locally, keep npx convex dev running in a separate frontend terminal while Vite runs.

API Smoke Test

Health:

curl http://127.0.0.1:8080/health

Start an optimization job:

JOB_ID=$(curl -s -X POST http://127.0.0.1:8080/optimize \
  -H 'Content-Type: application/json' \
  -d '{
    "url":"https://example.com",
    "target_customer":"startup founder",
    "goal":"increase signups",
    "iterations":1,
    "constraints":{
      "brand":{"colors":["#0D7D59"],"fonts":["Inter"],"tone":"clear","logo_present":false},
      "locked_elements":[],
      "aggressiveness":"balanced"
    }
  }' | python3 -c 'import sys,json; print(json.load(sys.stdin)["job_id"])')

curl -N "http://127.0.0.1:8080/job/$JOB_ID/stream"
curl -s "http://127.0.0.1:8080/job/$JOB_ID/result" | python3 -m json.tool
open "http://127.0.0.1:8080/job/$JOB_ID/heatmap"
open "http://127.0.0.1:8080/job/$JOB_ID/variant/1/image"

Main Endpoints

• GET /health
• POST /capture
• GET /capture/{capture_id}/image
• GET /capture/{capture_id}/heatmap
• POST /score/{capture_id}
• POST /optimize
• GET /job/{job_id}/stream
• GET /job/{job_id}/result
• GET /job/{job_id}/heatmap
• GET /job/{job_id}/variant/{iteration}/image
• GET /job/{job_id}/best-image

Notes

• A run can take 30-90 seconds because OpenAI image editing is slower than text calls.
• If OpenAI text, vision, or scoring calls fail, the backend uses local fallbacks so development does not crash. If image editing fails, Fixate reports the failure instead of creating a fake overlay image.
• Do not commit backend/.env; it contains your API key.