Drop-in security gateway for AI agents: prompt-injection defense, PII scanning, tamper-evident audit. Gate everything. Trust nothing.
- Tweets 56
- Following 24
- Followers 554
- Likes 10
Gate AI affiliates earn 30% of Pro revenue for the first 12 months, then 15% for as long as the account stays subscribed. Gate takes no card at signup, so referrals are written onto the account and survive the trial.
constellationgate.ai/affilia…
3
9
29
997
New on the Gate AI blog: CEO @BenJorgensen on the trade we’re making with AI, the need for real accountability, and why its future shouldn’t be decided for us and explained later.
My original op-ed on the future of AI. There are trade offs and this isn't the panacea for everyone. @aiaiholdings @Conste11ation
From a macro perspective, there is no plan for AI - it's on everyone to start building that plan and putting in controls. constellationgate.ai/blog/th…
1
6
40
1,533
Gate didn’t start as a security product. It started as Swarm Deck an internal agent swarm the protocol team built because they were burning AI budget debugging the Hypergraph.
Then they had to lock the swarm down.
That’s how the benchmark paper happened.
piped.video/FLauzLpBwXY?si=6v0a…
2
13
29
8,488
Gate AI retweeted
As it turns out, Ox Alpha is GLM 5.3 Flash!
Try it out on @_GateAI
2
13
587
When it rains, it pours. Another great model release today: Qwen3.8-Flash.
Near-frontier performance across coding and agent benchmarks at up to 50x lower cost than Opus 5.
Multimodal, 1M context, and now live on Constellation Gate.
⚡Meet Qwen3.8-Flash, a multimodal MoE and an early preview of the Qwen4 architecture, now open-weight!
The production version Qwen3.8-Flash will be available soon via QwenCloud API at just $ 0.16/1M input tokens and $ 0.47/1M output tokens.
125B parameters + 51B N-gram embeddings, with just 6B activated per token. Unmatched cost-efficiency.
What's new: 🥳
- Next architecture: GDN + QSA hybrid attention, Gated Residual, N-gram Embedding & Muon optimizer, serving as a precursor to the architecture used in Qwen4.
- Dramatically lower training and inference costs: trained at just 1/9 the cost of Qwen3.7-Plus, while outperforming it across the board with especially strong gains in coding and office tasks.
- Strong performance: scoring 58.7 on DeepSWE 1.1, 62.5 on SWE-bench Pro, 73.9 on CoWorkBench, 84.5 on AndroidWorld, and 95.7 on MathVision (with CI).
- 262K native context, extensible to 1M with YaRN.
We’re also releasing the weights for Qwen3.8-Flash-Next, giving the community an early look at the new architecture we’re exploring for Qwen4.🚀
We can't wait to see what you build with Qwen3.8-Flash!👀👇
- Blog: qwen.ai/blog?id=qwen3.8-flas…
- Technical Report: github.com/QwenLM/Qwen3.8-Fl…
- Hugging Face: huggingface.co/Qwen/Qwen3.8-…
- ModelScope: modelscope.cn/models/Qwen/Qw…
3
8
38
1,102
Mystery solved: Ox Alpha is GLM-5.3-Flash from Z.ai.
Near-frontier intelligence at up to 50x lower cost than Opus 5.
And it’s now live on Gate.
Try GLM-5.3-Flash today.
Introducing GLM-5.3-Flash
- Leading capabilities at a highly competitive price
- Natively multimodal with a 1M-token context window
- A 320B-A18B model released under the MIT License
- Previously previewed as Ox Alpha, running entirely on Chinese AI chips
Blog: z.ai/blog/glm-5.3-flash
Available now across all official platforms:
Weights: huggingface.co/zai-org/GLM-5…
API: docs.z.ai/guides/llm/glm-5.3…
Coding Plan: z.ai/subscribe
ZCode: zcode.z.ai/en
Chat: chat.z.ai
AutoClaw: autoclaw.z.ai
2
9
22
1,116
New: Gatekeeper. Help built into the Gate dashboard. Ask how anything works, or ask about your own account state. Answers without leaving the app, and never reads a prompt.
app.constellationgate.ai
ALT The Gate AI dashboard with the Gatekeeper help panel open on the right. The panel greets the user with "How can I assist you today?" and offers four suggestions: learn the basics of Gate AI, set up the Gate Connect app, add team members, and contact support. Behind it, the dashboard shows weekly metrics: 651 messages, 16% tokens saved, and 3 threats detected.
9
27
4,905
74% of enterprises believe they could pass an AI compliance audit today. 27% call their governance fully mature.
Separately, 71% say they can produce a defensible audit trail against 19% who have the logging and retention controls to do it. Both surveys came from vendors. The gaps did not.
New Gate AI blog. Delaware Chancery quoted a CEO's deleted ChatGPT logs in a $250M earnout case. The substance came back through ordinary discovery. A record assembled by the other side is never the one you would have written. Keep your own.
constellationgate.ai/blog/de…
4
21
672
New on the blog: our CTO @codebrandes on software that rewrites itself after you install it, and what that means for anyone trying to keep AI under control.
constellationgate.ai/blog/so…
Software is writing itself, and everything we thought we knew about securing it assumed a program was a fixed thing.
I've been chewing on this since I bought a $40 dev board and realized I hadn't bought a product. I'd bought something waiting to be told what it was.
constellationgate.ai/blog/so…
1
3
25
1,298
New benchmark: multi-turn prompt injection against 14 frontier models.
Most were vulnerable including GPT Sol, Grok 4.6, and Kimi K3, and Claude Sonnet. Claude Opus and Fable performed the best.
Full results on the blog:
constellationgate.ai/blog/fr…
Turns out you don’t need fancy techniques to break the latest models. GPT 5.6 Sol and many other frontier models fall even to relatively simple attacks.
constellationgate.ai/blog/fr…
3
11
54
326,125
Gate screens the request before it reaches any model, so the verdict does not depend on the model. We replayed the attacks with the policy set to Block: blocked 4/4 in each.
Reproduction scripts:
github.com/Constellation-Lab…
1
6
208
Rocket Resume takes roughly 150,000 visits a month and nearly every session invokes a language model. The token bill was about $40,000 a month.
Inline caching and prompt compression cut it 23% with no application code changes, verified per workspace against the real invoices.
9
44
756
Gate AI retweeted
First public customer numbers for Gate AI. ~$40K/month on AI tokens, 23% daily savings without code changes, and every request landing on a tamper-evident record. Compression pays for the platform. Audit is the tier that keeps paying after that.
Rocket Resume was spending ~$40K a month on AI tokens. They routed traffic through Gate AI.
Caching + prompt compression delivered ~23% daily savings on the same workload. Every request now lands on a tamper-evident audit log.
Drop-in. No code changes. Full visibility for the team.
constellationgate.ai/custome…
2
5
28
653
Rocket Resume was spending ~$40K a month on AI tokens. They routed traffic through Gate AI.
Caching + prompt compression delivered ~23% daily savings on the same workload. Every request now lands on a tamper-evident audit log.
Drop-in. No code changes. Full visibility for the team.
constellationgate.ai/custome…
5
17
50
10,412
Every AI workflow now carries three costs: the tokens it burns, the human work it reshapes, and the governance to explain the result later. One arrives as an invoice. The other two arrive as a surprise.
constellationgate.ai/blog/ai…
1
11
33
5,515
