Let agents operateyour IBM i.
The people who know the green screens are retiring, and most of what they know was never written down. Tower lets an agent do the keystrokes on the screens you already run, with a person’s approval where it matters and a signed log of every entry. We build the connection to your system with you.
One workflow, 90 days, your real system, 1,000 live writes included. The adapter runs on your side and connects outbound; Tower never needs a way in. After the pilot, $0.25 per executed write.
Nothing is billed during a pilot without a conversation first.
Two ways to start
A Tower pilot
We pick one workflow with you, order entry or something as contained, and build the adapter that connects it to your IBM i, SAP or anything with a green screen. You start in a test environment and move to production when your own team says so. What we need: one person who owns the workflow, access to a test or QA environment, and the three or four operations the agent should be allowed to call.
Agent-ready, done for you
We make your product findable, readable and usable by agents: llms.txt, markdown pages, OpenAPI discovery, a hosted MCP server and a Registry listing. A fixed price from $1,500, delivered as pull requests, measured by the same check before and after.
How Tower decides
Call. Decide. Act.
Agents get blocked for not identifying themselves, or trusted with writes nobody can audit. Tower is the third option: every write is a decision, every decision has a reason, and every step has a receipt someone can verify later. First target: IBM i order entry at mid-market distributors.
The safety layer · Guardrails
Every answer, checked before it ships.
Inside Tower, and on its own: send any model output and get back a verdict, rule by rule, with calibrated confidence and one recommended action (block, redact, review or allow) in about 150 milliseconds. It knows who the message is going to, so a customer’s own phone number is not a leak and another customer’s is. 10 checks a day with no key. The send gate.
Live · no signup
POST /api/v1/evaluate · content-safety
# request
curl -X POST https://overwing.ai/api/v1/evaluate \
-H "Authorization: Bearer ow_live_..." \
-H "Content-Type: application/json" \
-d '{"input": "<model output>", "rule_set": "outbound-message",
"context": {"recipient": "customer", "channel": "email"}}'# response
{
"verdict": "fail",
"recommended_action": "redact",
"confidence": 0.96,
"latency_ms": 148,
"results": [
{ "rule": "unauthorized_pii", "answer": true,
"confidence": 0.96, "verdict": "fail", "action": "redact" },
{ "rule": "toxicity", "answer": "safe",
"confidence": 1.0, "verdict": "pass", "action": "block" }
]
}What Overwing makes
Five products. One shape: a check, before the act.
Overwing Tower
Clearance for agents that operate legacy business systems. Each write is executed, sent to a person for review, or refused, and every step gets a signed receipt. In pilot, starting with IBM i order entry.
Read about TowerOverwing Guardrails
Checks text before an agent sends it or acts on it. One call returns a verdict of pass, fail or review, one recommended action, and a result for each rule with calibrated confidence. The safety layer inside Tower, and on its own.
How it worksOverwing Atlas
Open data on AI agents. A registry of 350 crawlers, fetchers and browser agents, a lookup that says what a User-Agent claims to be and whether the claim can be verified, first-hand scans of whole sectors, and a weekly note on agent traffic.
Explore AtlasOverwing Beacon
Is your product reachable by agents? One lookup of a URL says whether agents can find it, read it and use it: llms.txt, MCP, A2A, OpenAPI and how the page reads to a model. Free.
Check your siteOverwing Preflight
Checks a Solana transaction, or an x402 payment, before an agent signs it. One call works out what it would take from the wallet, or whether the payee and the price are what they claim, and answers allow, review or refuse against limits you set.
Read about PreflightAll five answer over a REST API. Every one but Preflight also answers over a hosted MCP server at overwing.ai/mcp, the A2A protocol, and SDKs for JavaScript and Python. 10 evaluations and 10 lookups a day need no key.
The whole guide in plain textA guardrail is a signal, not a verdict you defer to.
- Sep 22 Guardrails live
- Sep 24 x402 settles on Base
- Sep 25 Atlas registry
- Sep 28 Tower pilot
- Sep 30 Beacon
- Oct 8 Preflight for Solana
- Oct 13 Preflight for x402
Why these guardrails
Score-only guardrails vs Overwing
| Capability | Score-only tools | Overwing |
|---|---|---|
| What you get back | A score, and a threshold you invent | A verdict, per-rule answers, and one recommended action |
| Uncertainty | Rounded away | An explicit review state, routed to a person |
| Context | The text alone | Recipient, channel, and ownership, so the customer's own phone number is not a leak |
| Rules | A fixed list | Plain-language questions you write, with actions and weights |
| Paying | A card and a sales call | Card, USDC at checkout, or per call over x402 with no account |
| Who can use it | Humans with dashboards | Humans, and the agents they run, all as JSON |
Pricing
Per organization. Cancel any time.
Limits are evaluations per UTC day, shared by all your keys. Every plan has every endpoint, custom rule sets, webhooks, and the dashboard. Everything on one page.
Try every endpoint. Enough to wire up a prototype or run an agent's smoke tests.
Start freeFor a single product or agent in production.
Choose StarterFor teams and fleets of agents with steady traffic.
Choose GrowthHigh-volume pipelines. Talk to us above this.
Choose ScaleTop up by card, no subscription. 2,000 evaluations for $5, up to 1,000,000 for $999. Credits never expire and are spent after your daily allowance. Create an account and buy on the billing page.
Done for you. We make your product findable, readable and usable by agents for a fixed price from $1,500, delivered as pull requests. Tower clearance for agents on legacy systems is $0.25 per executed action. All prices.
Agents with a wallet. Every paid route answers an unpaid request with a 402 and its terms, in USDC on Base over x402. No account. Machine-readable at /api/v1/plans.
Overwing Atlas
Atlas. Open data on AI agents.
Who the agents are, where they go, what they spend. A registry of 350 crawlers, fetchers, and browser agents with verification methods and signing keys, a lookup API that says what a User-Agent string claims to be and whether the claim can be trusted, and first-hand research on agent traffic and spending.
Should the agent sign this?
An agent with a wallet signs whatever it is handed. Preflight is one call before the signature. For Solana it simulates the unsigned transaction and checks what it would take from your wallet against limits you set; a verdict covers a transaction that lands within 120 seconds, and every verdict is published. For x402 it reads the payment terms and checks the payee against the directory and the price against the record. One check counts as one evaluation, or $0.01 over x402 with no account.
Is your product reachable by agents?
Enter a URL. Beacon looks for what an agent looks for, an llms.txt, an MCP server, an A2A agent card, an OpenAPI document, a page it can read without a browser, and answers three questions: can an agent find you, read you, and use you. You get a score and the changes worth making first. Free: the full report opens with a free account, and an agent with a wallet and no account pays $1 over x402.
Agent-ready, as pull requests.
Your product, made findable, readable and usable by agents: llms.txt, markdown pages, OpenAPI discovery, a hosted MCP server and a Registry listing, delivered as pull requests your team merges. A fixed price from $1,500, measured by Beacon before and after. Your data partners and your customers’ agents are the ones who notice.
Agents welcome
If you are an agent, nothing here needs a human.
Sign up, pay, evaluate, rotate keys, and cancel, all as JSON. Errors name the field. Rate limits come back as headers with a reset time. Idempotency keys, batch evaluation, an OpenAPI schema, a hosted MCP server, an A2A endpoint, and the whole guide in plain text at /llms.txt. We want your traffic. The only ask is that you pay for what you use.
# a key, one call, no browser
curl -X POST https://overwing.ai/api/v1/signup \
-d '{"email":"bot@acme.dev","password":"…"}'
# or no account at all: pay per call in USDC over x402
curl -X POST https://overwing.ai/api/x402/evaluate \
-d '{"input":"…"}' # 402 → sign → resend
# MCP, for Claude, Cursor, and friends
https://overwing.ai/mcp # hosted, no install
npx overwing-mcp # or local; registry: ai.overwing/mcp
# A2A: send a message, get a verdict
POST https://overwing.ai/a2a # card: /.well-known/agent-card.jsonQuestions
Overwing, in plain answers.
- What is Overwing?
- Overwing makes five products for software that acts on its own. Overwing Guardrails checks text before an agent sends it or acts on it. Overwing Preflight checks a Solana transaction before an agent signs it. Overwing Atlas is open data on AI crawlers and agents, with a lookup that identifies one from its User-Agent. Overwing Beacon says whether a product is reachable by agents. Overwing Tower gives agents clearance to operate legacy business systems. All five work over plain HTTPS and JSON, and an agent can sign up, pay and cancel without a browser.
- What is Overwing Guardrails?
- Guardrails is an API that checks text: a model's output, a draft message, or something received. It returns a verdict of pass, fail or review, one recommended action of block, redact, review or allow, and a result for each rule with a calibrated confidence. It checks for personal data, confidential leaks, toxicity, self-harm and sexual content, and you can write your own rules in plain language. 250 evaluations a day are free with a key, and 10 a day need no key at all.
- What is Overwing Atlas?
- Atlas is open data on AI agents: a registry of 350 crawlers, fetchers and browser agents with their operators, purposes and verification methods; a lookup that says what a User-Agent string claims to be and whether that claim can be verified; and first-hand scans of how whole sectors treat agents. 10 lookups a day need no key.
- What is Overwing Preflight?
- Preflight answers one question before an agent signs a Solana transaction: should it? One call simulates the unsigned transaction against the chain as it is now, works out what it would take from the wallet you name in SOL, tokens and control of its token accounts, and answers allow or refuse against limits you set, with every reason. It never sees a private key. A verdict describes one moment and covers a transaction that lands within two minutes. Every verdict is published. One check counts as one evaluation on any plan, or $0.01 over x402 with no account.
- What is Overwing Beacon?
- Beacon answers one question about a site: is this product reachable by agents? It checks whether an agent can find it (robots.txt rules for AI agents, llms.txt, a sitemap, the MCP Registry), read it (whether the page is served to automated clients, whether it has content without JavaScript, whether it says what the product does) and use it (an MCP server card and hosted endpoint, an A2A agent card, an OpenAPI document). It returns a score, the three answers, and the changes worth making first. A check is free: anyone gets the score, the three answers and the first fix, and the full report opens with a free Overwing account. An agent with no account can pay $1 in USDC over x402 instead. There is no subscription, and a sample report is at https://overwing.ai/beacon.
- What is Overwing Tower?
- Tower is a decision, permission and receipt layer that lets AI agents operate legacy business systems, starting with order entry on IBM i (AS/400). Agents call typed operations with their own scoped keys. Tower executes each write, sends it to a person for review, or refuses it, and signs a tamper-evident receipt for every step. Tower is in pilot: the order-entry workflow runs against a mock adapter, and adapters for real systems are built per pilot. It costs $0.25 per executed action, with 1,000 decisions and dry runs free each month.
- How does an AI agent connect to Overwing?
- Five ways. The REST API, described at https://overwing.ai/api/v1/openapi.json. A hosted MCP server at https://overwing.ai/mcp, with no install. The same MCP server from npm: npx -y overwing-mcp. The A2A protocol at https://overwing.ai/a2a. And SDKs for JavaScript and Python, both named overwing. An agent with a wallet can also pay $0.002 per evaluation in USDC over x402 with no account. The full guide is at https://overwing.ai/llms.txt.
- Does Overwing work on text that is not in English?
- Yes. The input can be in any language the model reads. It has been tested in Spanish, Portuguese, French, German, Japanese, Chinese (Simplified), Korean, Arabic and Hindi; that was a small test, not a benchmark. Responses are in English.
Your first pilot is one email away.
2 free pilot places: one workflow, 90 days, your real system. Or start smaller: 10 guardrail checks a day with no key, 250 with a free account.