The Problem
AI‑agent teams spend most of their effort wiring together separate observability, evaluation, and guard‑rail services. The resulting toolchain is fragile, hard to version, and does not close the feedback loop needed for agents to self‑improve.
What This Does
Future‑AGI ships a single‑page React UI that aggregates tracing, evaluation, simulation, dataset management, a gateway, and guard‑rail dashboards. The UI lives under frontend/ and is built with the standard React toolchain (package.json, webpack/vite implied).
Key entry points:
frontend/.storybook/main.js – Storybook configuration for component development. frontend/index.html – HTML host for the React bundle. frontend/Dockerfile – Produces a container that serves the built UI via Nginx (frontend/nginx.conf).
The platform is intended to be self‑hosted; the root docker‑compose.yml orchestrates the UI container together with any backend services (not present in this fork). The .env.example files list required variables (e.g., NEXTPUBLICAPIURL) for the frontend to reach the API layer.
How To Use It
Setup – Install Node dependencies and build the UI:
cd frontend npm ci # uses package-lock.json npm run build # creates a production bundle in /dist (defined in package.json) Configuration – Copy the example env file and adjust the API endpoint:
cp .env.example .env # edit .env, e.g. set NEXTPUBLICAPIURL=https://my-backend.example.com Run locally (Docker) – The root compose file builds and runs the UI container:
docker compose up --build # uses docker-compose.yml which references frontend/Dockerfile
The UI will be reachable at http://localhost:3000 (port defined in frontend/nginx.conf). Deploy – CI pipelines in .github/workflows/ (e.g., frontend-deploy-eu.yaml, frontend-prod-deploy.yaml) already contain Docker image build and push steps; they can be adapted to a private registry.
No backend code is included, so a compatible API (the original Future‑AGI server) must be provisioned separately.
Real‑World Use
A product team can point NEXTPUBLICAPI_URL at a running Future‑AGI backend, then use the UI to:
Open a trace, expand a span, and click Run Eval → the UI posts to /evals/run and displays the score. Switch to Simulate → the UI streams a voice‑agent conversation from /simulate. View Gateway → cost and guard‑rail metrics are fetched from /gateway/metrics.
All interactions are driven by standard REST calls defined in the backend’s OpenAPI spec (outside this repo).
Code Health & Issues
Low – Missing backend – The repository contains only the frontend; the core tracing/eval services are absent, limiting immediate usefulness. Med – Limited test coverage – Only three test files are present; no CI step runs them (frontend-auto-approve-hotfix.yml focuses on lint/format). Low – Hard‑coded asset paths – UI references many static SVG/PNG files under frontend/public/assets/; any change in asset naming requires code updates. Low – No TypeScript – The codebase is pure JavaScript/JSX, which reduces static safety for a large UI. Low – Environment variable leakage risk – .env.example is present, but no validation of required vars is enforced in the UI startup script. Low – CI is UI‑focused – Workflows handle lint, format, and deployment but do not run unit or integration tests, leaving regressions unchecked.
Overall, the repo follows standard Node/React conventions, includes linting (.eslintrc.cjs, .prettierrc), and Docker best practices (multi‑stage build, .dockerignore). License (Apache‑2.0) and contribution docs are present.
The Bottom Line
Future‑AGI provides a polished React front‑end that consolidates agent observability and evaluation into a single dashboard, and it can be containerised and deployed via the supplied Docker files. However, the lack of backend code and sparse testing mean the repo is only a UI component of a larger system; teams must already have or be willing to provision the corresponding server side. It is suitable for organizations that already use the official Future‑AGI backend and need a ready‑made UI layer, but not for a standalone install.