GradeX

We got tired of “this refactor feels faster.” So we built GradeX — a CLI that runs parallel AI agents on your repo, benchmarks every patch, and only keeps changes that actually improve the number discover → baseline → optimize → dashboard Real terminal Real git worktrees Real gates Open source MIT. pip install gradex Watch it run (Demo)👇

Went quiet for a few days, owe you an explanation Honestly just needed to step back for a bit, recharge, and think clearly about the next phase instead of just grinding and tweeting noise Came back with fresh eyes on Cloud, cleaner priorities, and a better sense of what actually needs to ship next No drama, no issues, just a reset Back now and building harder than before

Small patch day - $GradeX - Cleaned up duplicate config loading on startup - Fixed an edge case where stall detection miscounted on the first round - Improved logging output for failed gates, easier to see why something broke - Minor dependency bump for the dashboard's websocket layer

GradeX doesn't just keep the first patch that passes Every PASSED experiment in a round gets compared against the current best score, not just the original baseline If experiment 3 passes but experiment 4 passes with a better number, experiment 4 becomes the new best The bar keeps moving up, never down, across an entire run

GradeX v0.2.0 is out pip install -U gradex - Drill into every experiment — click a row, see branch, score, gate, tokens, full trace timeline - gradex traces — see exactly what each experiment did, no more blind optimize loops - gradex stats — token counts and estimated $ per run - discover now supports Node/TypeScript repos, detects package.json, writes the benchmark, gates on npm test Python stays default. More languages coming Local dashboard today. Hosted team view is what Cloud is building toward

Going live on pump.fun right now Been deep in building mode the last few days, $GradeX Cloud planning, roadmap updates, fixes nobody sees in a tweet Figured instead of just writing about it, you should actually watch it happen Pulling up the real CLI, the real dashboard, no script, no slides pump.fun/coin/2njgPSCP7QBFff…

Every AI patch in GradeX ends in exactly one of three states - PASSED — beats baseline AND every test passes - REJECTED — tests broke, speed doesn't matter - FAILED — crash, timeout, or no real improvement No fourth outcome where something ships because it "sounds right" The gate doesn't care about confidence Only the number and the test suite

Shipped on GradeX 🚀 New in the CLI: → OpenRouter — try free-tier models with one API key → gradex report — shareable HTML proof of every run → Benchmark cache — faster re-runs during optimize → Interactive setup — gradex configure saves provider + model after install pip install -U gradex gradex install cursor gradex discover "make this repo faster" gradex optimize Open source: github.com/gradex1606-dev/Gr…

Dev Logs // GradeX Build Session Back at it today Today was about the bridge between local and cloud Focused on: 🔄 Run bundle sync 🔐 Authenticated cloud client 📊 Workspace + history endpoints ⚡ Live experiment streaming Building toward $GradeX Cloud ↓ Local runs stay on your machine. Upload a run bundle → team dashboard → live experiment stream. Build session attached🎥 Open source CLI today: pip install -U gradex Early access → gradex.fun

GradeX Cloud, a closer look Today, every dev runs GradeX alone, own laptop, own worktrees, own localhost dashboard Cloud changes that The same loop — discover, benchmark, optimize, gate, prove — moves to managed infrastructure. Sandboxes spin up remotely, agents run in parallel without anyone's machine doing the work, and the whole team watches one shared dashboard instead of five different localhosts History stays searchable across the whole org, not stuck on whoever's laptop ran it last This is what turns GradeX from a solo dev tool into something a whole engineering team can actually run on

Been quiet the last couple days, not gone anywhere, just heads down planning something bigger GradeX Cloud is coming in the next few days Same core loop — discover, benchmark, optimize, gate, prove — just hosted now - Live team dashboards, no more localhost only - Shared run history across the whole team - Managed sandbox workers, no local setup needed - Budget controls across providers - Full audit trail on every patch and gate CLI stays open source and free forever. Cloud is the part that needed real infra, so it needed real time Back to building. More soon

Sneak peek at what's running behind the scenes right now Phase 2, Intelligence Layer — 62% done, requires few more days to complete Multi-agent workflows, reasoning engine, advanced experiment generation, all merged 12/12 workers active, 3 PRs merged in the last 12 hours More soon

We opened early access on gradex.fun If you want updates as we ship new experiment and verification features before anyone else hears about them, get on the list Join Early Access 👇

Seeing a lot of people ask about paying for DEX right now Here's the actual plan — DEX update is scheduled around 50k mcap, that's the checkpoint we're building toward, not something we're rushing to pay for early Until then, we keep shipping New roadmap phases, Cloud groundwork, CLI improvements, all of it happening whether the chart is watching or not We didn't launch this for a quick flip, GradeX existed before the token did and it'll keep getting built after Long term. That's the only mode we're in

Where GradeX is going Today — agent-assisted engineering Parallel AI agents, benchmarks, gates. Every experiment measured against a baseline before anything merges. Tomorrow — agent-driven engineering Agents propose, test, and refine continuously. Teams review evidence backed results instead of debating untested ideas. Future — autonomous engineering systems Self-directed experimentation across repos, intelligent rollbacks, adaptive benchmarks that keep software improving on its own. We're at Today. Building toward Future.

Introducing token utility for $GradeX GradeX Cloud is coming — hosted dashboards, team run history, managed sandbox workers, shared budget controls, full audit trail on every patch and gate Access to Cloud will require GradeX tokens The CLI stays open source and free, that's not changing. Cloud is where the token comes in

$GradeX roadmap is live Phase 1, Foundation — shipped Core CLI, Experiment Engine, Benchmarking, Dashboard, Repository Management, Verification Pipeline Phase 2, Intelligence Layer — in progress Token util,Multi-agent workflows, reasoning engine, advanced experiment generation, better evaluation Phase 3, Team Collaboration — planned Shared workspaces, experiment sharing, team dashboards, role permissions Phase 4, Continuous Improvement — planned background optimization, automatic scheduling, long term learning Phase 5, Autonomous Engineering — future vision Self-directed experimentation, cross-repo learning, intelligent rollbacks

Here's what happens in one GradeX optimize round, step by step - Each subagent gets the same brief: current best score, what's already been tried, what failed - They each propose a patch independently, no knowledge of each other's attempts - Every patch runs in its own worktree, against the same benchmark, same gate - Whichever one passes becomes the new best. Everyone else is discarded No agent ever copies another's homework. They're all solving blind, in parallel

GradeX is built like production software, not a weekend hack - Results get cached same patch attempted twice doesn't re-run the benchmark, saves time, API calls, and money - Secrets get scrubbed before anything is logged or sent to a model - Path safety stops an agent from wandering outside its worktree - Rate limiting built in, so a runaway loop can't drain your API quota Efficiency and safety, both by design

Today's ship log - Cleaned up error handling on first run - Fixed a docs/CLI mismatch on provider list - Polished the demo repo so it's not empty anymore None of this is glamorous, all of it makes the tool trust itself a little more