Launching today
Constellation Gate AI
Prompt injection and token savings - #1 in benchmarks
130 followers
Prompt injection and token savings - #1 in benchmarks
130 followers
Point your AI agent at Gate. Inherit prompt-injection defense, secret scanning, and a verifiable audit trail. Gate includes prompt compression and caching so you can reduce token usage by 20-40% without changing model output. In AI security benchmarks, Gate ranks #1 across 16 public prompt-injection datasets. Keep your Claude or ChatGPT subscriptions and route them through Gate, or choose from 100+ models pay-as-you-go. Setup is easy: no code changes with our desktop app.
Constellation Gate AI
👋 Hey Product Hunt! I'm Ben from Constellation Network — excited to launch Gate AI today. I believe this is an incredible tool for both management of engineering teams to use as well as the individual developer. We made the platform extremely easy to onboard that even your CFO would understand it in minutes. This provides audibility around your organizations AI usage (while saving you money and providing security. Our goal is that Gate AI can become a standard tool across organizations.
If you're running Claude CoWork, Cursor, an OpenAI/Anthropic-based agent, or anything that calls a model API, Gate sits inline as a proxy and does three things automatically:
🛡️ Blocks prompt injection: ranked #1 across 16 public benchmarks (97.4% F1), beating the leading enterprise vendor 96.6% to 83.7% F1 head-to-head
🔒 Redacts secrets & PII before they leak out in a response
💸 Cuts token spend 20%+ via lossless compression and prompt-cache injection — with zero code changes
Every decision Gate makes is sealed to a blockchain-anchored, independently verifiable audit trail — so you're not just told it's safe, you can prove it.
Free tier, no credit card.
Would love feedback from anyone shipping agents in production and management looking for better AI controls!
genuine question on the "blockchain-anchored" audit trail claim - what does the blockchain actually buy you over a plain signed hash chain (like a Merkle tree you publish periodically)? tamper-evidence works fine with either, and blockchain usually only earns its cost when multiple mutually-distrusting parties need to verify something without a shared authority. for an audit trail that's presumably verified by the org itself or an auditor they already trust, that bar seems higher than what's needed here, unless I'm missing what it adds
Constellation Gate AI
@galdayan Great question. The blockchain anchoring is an automated way to publish Merkle roots of your data to a public ledger. The advantage over a self-published chain is that the timestamp is fixed externally - we can prove the state of your data at each point in time and that it wasn't altered afterward, precisely because we don't control the anchor. A timestamp you set yourself doesn't prove anything, but one anchored to a public ledger does.
It arguably exceeds the standard that would be demanded by an auditor but we believe it is a more complete solution for tamper-evident logs. The cost is highly optimized and not really a factor - it's free for all account types.
How does the immutable audit trail actually work in practice, is there a per-call cost or does it come bundled with the free tier?
Constellation Gate AI
@satanuurscfq The audit trail is bundled with the free tier. The audits leverage Constellation's Digital Evidence technology. In the future, we'll add more reporting features that may end up in paid tiers. The basic auditing will remain free.
Curious what the benchmark setup looks like: are you measuring against known jailbreak/prompt injection datasets, real agent workflows, or synthetic prompts? Also wondering whether the token savings come from prompt compression, routing, filtering, or something else under the hood.
Constellation Gate AI
@crystalmei Good questions. Our published benchmarks measure against 16 public datasets covering both known injection/attack scenarios and benign prompts. We train against real agent attack scenarios as well but focused on datasets in the benchmarks that allow comparison against other systems. You can read more about the benchmark methodology on our blog post or read the arXiv report if you want to deep dive on the specifics.
The tokens savings come from a combination of inline prompt compression and response caching. Most workflows see 20-30% token savings from the compression alone. Response caching is configurable and more dependent on your workflow but can save a significant additional amount on repetitive workflows.
Combining prompt injection protection with token optimization is an interesting mix since people usually think about those separately. Have you found that customers come for the security side first, or are the benchmark results what usually gets their attention?
Constellation Gate AI
@amjad_shaik Honestly a bit of both, but the token savings tend to be the bigger draw right now - especially with Fable burning through tokens on Claude subscriptions so fast lately.
The prompt injection security is the harder feature for other platforms to match, and nearly all the current options are enterprise-only. We wanted something anyone could sign up for to protect their workflows without booking a sales call.
Overall we want the platform to be a place anyone running AI agents can plug into for visibility into what their agents are doing, protection against prompt injection and other attacks, and lower costs, without giving up the tools they're already using.
@codebrandes Thanks for taking the time to explain it. I hadn’t realized token optimization was currently a bigger adoption driver than security for many teams. The idea of adding protection without forcing teams to replace their existing stack is a practical approach.
Constellation Gate AI
Hi @product . Dave here, CPO at Constellation Network. @Gate AI is what we've been building since March.
Prompt injection is OWASP's #1 LLM threat. The commercial defenses I could recommend to a friend a year ago all started at six figures. If you're one dev running a coding agent, the honest answer has been "figure it out yourself." That's the gap we built Gate AI to close.
Point your AI agent at Gate. No code changes. Inherit prompt-injection defense (F1 97.4% at 1% FPR across 16 public benchmarks; full methodology on arXiv at 2606.02959, we published it so the numbers can be audited), secret and PII scanning, and an audit trail sealed to Constellation's Digital Evidence layer that a regulator can verify without trusting us. Compression saves 20%+ tokens automatically. Early access pro users are running closer to 30%+.
Free tier available, no card. Paid and pay-as-you-go for teams and heavier usage.
What I most want feedback on: does the zero-code onboarding hold for your setup? Does the audit trail cover what your compliance team asks for? If something is confusing, tell me. We'll fix it fast.
"#1 across 16 public prompt-injection datasets" is a strong claim - is that benchmark run by you internally or is there a third party leaderboard I can check that number against? also the 20-40% token savings "without changing model output" - does the compression step ever get caught out by a prompt that relies on exact wording (like a few-shot example)
Constellation Gate AI
@omri_ben_shoham1 The benchmarks were run by our team and we published the complete methodology so that anyone can replicate the results. You can check out the blog post or arXiv report if you want to dig into the specifics.
The prompt compression never changes prompt wording - it saves tokens by removing whitespace and invisible tokens, and reformatting inefficient or duplicate tool results. There are specific strategies for different tools that are designed to optimize for token count without modifying the information available to the model.