ProxyWatch
Ready nowv0.3.0MIT
Watches a protocol's upgrade and admin surface over public RPC and reports the moment it changes, with severity and on-chain evidence.
Details · $29/mo self-serve
Read-only · No keys · Public RPC
Know what can change your contracts,
and who can change it.
Two questions decide how much a compromised key can do: can this contract's logic be replaced, and who holds that power? These tools answer both — from public chain data, without you handing over a key or an integration.
Every alert names the block number and transaction hash behind it. If a claim can't be checked on-chain, it isn't on this site.
Readiness — stated plainly
One product is ready today: ProxyWatch. It is built, its test suite passes, and it is live-verified on Ethereum and Base.
FundTrace is early access — it works end to end, but it is pilot-grade and quoted per job. AgentWire is not ready to sell: the API and feed are real, but the incident record behind them is still too small to be worth paying for. Rather than present three equals, we label them.
Products
Three working tools. Each answers one narrow question, and each answer can be checked against the chain.
FundTrace
Early accessCLI
Takes one address and a block range, then reports every ERC-20 transfer it was party to, ranked counterparties, token totals, and simple flow patterns worth a second look.
Details · $99 per triage
AgentWire
Not readyPreview
A read-only HTTP API and Atom feed over a curated record of AI-agent incidents, with citations to each original source.
Details · $9–29/mo
What the products share
The same constraints apply across the portfolio. They are design decisions, not marketing lines.
- Read-only, by construction. The only calls made are
eth_getStorageAt,eth_getCode,eth_call,eth_getLogsandeth_blockNumber. There is no code path that signs or submits. - No keys, so nothing to leak. A leaked context cannot move funds, because none exists. That removes a class of risk instead of managing it.
- Public endpoints with failover. No paid data provider is required. Deep history needs an archive endpoint, and we say so rather than silently returning nothing.
- Verifiable by default. ProxyWatch alerts cite the block and transaction; FundTrace rows carry the transaction hash; the detector behind both is MIT and readable.
- Honest about misses. This is monitoring, not insurance. When a read fails we say so — we don't quietly report "no change".
Where the code lives
This site describes working software. The source sits under products/ in the repository.
| Product | Path | Reference |
|---|---|---|
| ProxyWatch | products/proxywatch | README.md, SPEC.md |
| FundTrace | products/fundtrace | README.md, docs/methodology.md |
| AgentWire | products/agentwire | README.md, docs/API.md |
Open source & research
The tools here are auditable, and the claims behind them are checkable.
proxywatch-core
The control-surface detector this portfolio is built on, as a standalone command you can run yourself. No API keys, no accounts, ~200 lines you can read in full.
Read the source · MIT
Proxy admin prevalence
In our sample of 1,184 proxies across Ethereum and Base, 6.4% of those with an identified admin still used a single externally-owned account. The finding was concentrated in one proxy pattern and moved sharply with sampling — it should not be generalised across proxy types. Method, counts and caveats included.
Reproducible · read the numbers
We publish the method and the limits alongside every number. If a claim can't be checked, it doesn't belong here.
Questions a buyer actually asks
Is this an audit?
No. It is monitoring and analysis. It does not review your code, and it is not a substitute for one.
Do you need keys or access?
No. Everything is read from public chain data. We never need a key, a signature, or access to your infrastructure.
How is severity decided?
For ProxyWatch: critical is an upgrade or admin/owner change, or an EOA holding upgrade authority; warning is any other control-surface change; info is baseline and no-change. The thresholds are stated, not implied.
What if you miss something?
We will not pretend otherwise: this is monitoring, not insurance, and no detector catches everything. What we do guarantee is that a failed read is reported as a failure, never as "all clear".
What do you store?
The contract addresses and ranges you ask us to watch, and the public chain data we read. Nothing about your infrastructure.
How do I stop paying?
Month to month, no contract, cancel any time. Invoiced in crypto — no card, no KYC, no account required beyond the address we watch.
Limits
What this portfolio is not
These tools are monitoring and analysis software. They are not audits, not insurance, and not legal, financial, or investment advice. They do not prevent losses, remediate incidents, or recover funds, and they carry no guarantee that a given change or transaction will be detected.
FundTrace and AgentWire report heuristics and records that can be wrong in both directions. Treat any output as a starting point to verify, not a conclusion. Each product page states its own limits in more detail.