The SuperQwen MVP is complete. ✅
Under two weeks from the first commit to a live AI coding agent that plans, edits a sandboxed copy, and proves its work with tests.
Every sprint on the roadmap, shipped in public.
🚧 Sprint 5: in progress.
Sprint 4 made Cue safe. Now we measure how well it works, with numbers anyone can rerun.
First run: 6 of 8 patches validated and genuinely fixing the goal.
Roadmap: https://t.co/d1y3GXjGhU
Next, Sprint 5: benchmark & pilot.
Sprint 5 plan:
→ A repeatable coding evaluation set
→ Real success-rate, latency and resource numbers
→ A small developer pilot.
Follow along at https://t.co/d1y3GXjGhU
Sprint 4 is done 🎉🎉
We ran a real task on https://t.co/llAYQruKNs and read the sandbox's own evidence: validated with no network, under the task's own uid.
Policies, approvals and audit records now apply to every task.
What the test checks:
→ Nothing is written outside the task's copy
→ The real checkout is untouched
→ The planted secret never reaches a stored output, artifact or prompt
→ A planted test that tries to send the environment out can't open a socket
Sprint 4: we told our test model to obey an attacker.
→ Isolation and prompt-injection tests ⏳
A repo hides instructions for AI agents and a planted .env. The model follows all of them: read the secret, upload it, force-push, plant a git hook.
Every attempt was refused and audited.
Also new: known credential formats are redacted from stored output before the model sees it again, and a finished task's working copy is deleted after 14 days.
The owner of a task can read its full audit trail.
Sprint 4, day 2: security & recovery
→ Sandbox hardening and audit logs ✅
When the agent tests its patch, that code can't reach the network, even on a host without network namespaces: a seccomp filter only allows local Unix sockets.
Each task also runs as its own user.
Conviction is the part most people skip.
Ours: AI agents will write a lot of the world's code, so that code has to come with proof. That's why we built SuperQwen to plan first, wait for approval, and only call it done when tests pass.
One week in, and it's live.
I believe @PalmerLuckey built Oculus and Anduril with conviction about where the world was headed. I think the same thing is happening with the team @EreborBank, and it's now part of the RVI portfolio.
Sprint 4 is in progress: security & recovery
→ Policy engine and approval records ✅
Every tool call the agent makes is checked against the workspace's policy, which can only make it stricter. Every decision, allow included, is recorded.
$46.3K in 24h volume on $SQWEN, tracked on GMGN. 🔵
A week ago SuperQwen was an empty repo. Today it's a live AI coding agent with a community behind it.
To every trader and holder: thank you. We're proud of what we're building, and we'll keep shipping 🎉🎉
SuperQwen, week 1 recap 🔵
Shipped:
→ live site
→ Cue edits code for everyone
→ Public task log 52 commits · 514 tests passing
→ 4 open models
→ Best at 97.5% right next tool 10 tasks from 8 visitors
→ $SQWEN live on Pons
Next: Sprint 4, security ⏳
Sprint 4 is underway: security and recovery.
Sprints 1–3 gave SuperQwen an agent that reads your repo, plans, patches and proves it with your tests. Now we're making it safe to run on code you don't trust.
The dev team is building it now. What's coming:
Robinhood Chain
built for: the builders
built for: the apps
built for: the memes
built for: the wallets
built for: the trenches
built for: the LPs
built for: the markets
built for: the market makers
built for: Stock Tokens
built for: any asset
built for: onchain
built for: _______
$SQWEN is live on @RobinhoodApp Chain, launched via Pons.
Most tokens on a new chain are a name and a chart. $SQWEN comes with something you can use today.
SuperQwen is a self-hosted AI coding agent that has to show its work. Give it a goal at https://t.co/HbzcYdIcut: it reads the repo, drafts a plan you approve, edits a sandboxed copy, and only calls it done when tests pass. Every task is in a public log, and all four of our fine-tuned models are open on Hugging Face under Apache-2.0.
The product runs on our own servers. $SQWEN is the community token for the people following the build, on an Ethereum-settled L2 built on Arbitrum.
Robinhood Chain, meet a real AI product.
Not financial advice. Verify the CA only from this post or the pin on @SuperQwen_llm.
$SQWEN × SuperQwen, status update:
Token: live on Pons · DEX paid · 23.1M burned
Product: Cue edits code live · 4 open models · public task log
Next: Sprint 4, security
The token is live. The build keeps going.
CA pinned · NFA
🔥 23.1M $SQWEN burned.
Sent to the dead address. No one holds that key, so they're gone for good.
Don't take our word for it: check the transaction yourself. NFA.
Why it matters: on DexScreener anyone can launch a token called $SQWEN. Ours is the one with this profile and this contract address.
And the product keeps shipping: try the agent free at https://t.co/llAYQruKNs.
DEX paid ✅
The official $SQWEN profile is live on DexScreener: logo, banner, https://t.co/llAYQruKNs and @SuperQwen_llm, all in one place.
CA: 0xf8393a2f960eaf758df23d16f0e7a4b9be33d790
Always match the CA before you trade; lookalikes copy names, not addresses. NFA.