Agent-orchestrated reconnaissance
Reconnaissance you don't have to orchestrate.
Name a target. Zero-Recon drafts the plan, waits for your approval, runs the tools inside sandboxed containers under authorization you have proved, and returns one entity graph and one report — with whatever it could not check named, not scored around.
For security engineers, penetration testers, and bug bounty hunters who run this chain by hand today. Passive reconnaissance starts on the free daily allowance — no card, no authorization paperwork.
acme-demo.example
Passivestandard_reconMap the external surface and flag exposed services.
Surface discovery
subfinder · crt.sh · certspotter
31 subdomains
Resolution
dnsx · cdncheck
24 resolved · 3 behind a CDN
Service fingerprinting
httpx
18 live hosts
TLS inspection
tlsx
6 of 18 hosts timed out
Historical surface
wayback
not reached — mission ended first
Coverage 8 / 10 — security score withheld
TLS inspection is incomplete and historical surface was never collected. The report states both gaps instead of scoring around them.
The problem
The tools are not the hard part.
subfinder, dnsx, httpx and nuclei are each a single command. What costs the afternoon is everything between them.
- 01
You are the scheduler
Which tool runs next, against which slice of the last tool's output, and which steps have to be replayed now that the surface changed.
- 02
Nine outputs, nine shapes
Correlating subdomains, resolutions, certificates and HTTP fingerprints into one picture of the target is manual work, and it is where things get dropped.
- 03
Silence looks like safety
A tool that timed out and a tool that found nothing return the same empty result. A report assembled by hand inherits that ambiguity without flagging it.
- 04
Authorization lives in your head
Nothing in a local tool chain stops an active scan against a host you were never cleared to touch.
Zero-Recon takes all four and makes them the platform's responsibility: the ordering, the correlation, the distinction between “checked” and “never checked”, and the gate in front of anything that touches the target.
How it works
Four steps, start to report.
The same loop runs whether the mission takes ninety seconds or an hour.
State the target
One line of natural language in the workspace. The agent decides whether it is a reconnaissance request at all, and scopes a mission if it is.
Anything that is not a recon request is answered as an ordinary chat message instead.
Review the plan before it runs
You get the phases, the tool behind each step, the reasoning, and the estimated cost. Approve it, reject it, or describe a change and get a revised plan.
Proposed plan
est. 34 coins- 01Enumerate subdomainssubfinder
- 02Resolve and dedupednsx
- 03Fingerprint live hostshttpx
Tools execute under authorization
Each task runs in its own isolated container. Passive tools run freely; anything active is refused unless a signed authorization record covers the target.
Read a report that states its gaps
Findings, the entity graph, and a coverage line counted from the plan's own task statuses — which decides whether a score is published at all.
Mission report
md · jsonScore withheld — 2 of 10 planned steps did not complete.
The coverage rule
A finished-looking report can still be a lie.
When a mission stops halfway, the steps that never ran produce no findings. To a scoring engine that is indistinguishable from steps that ran and found nothing — so the silence gets inverted into a clean posture score.
Zero-Recon counts coverage first, from the plan's own task statuses, and gates every downstream conclusion on it: the score, the risk band, the summary wording, the recommendations. Below full coverage the report states what it does not know instead of scoring what it never measured.
The count is arithmetic, not generation. No language model participates in it, so it cannot fail open when a provider is down — and it is re-applied on every read, because a report written before the gate existed cannot be trusted to have passed through it.
Distinctions the platform refuses to collapse
- No findings is not the same as
- Clean
- Not executed is not the same as
- Passed
- Partial coverage is not the same as
- Low risk
- Tool timed out is not the same as
- Nothing there
Authorization & isolation
Passive by default. Active only where you can prove you are allowed.
Nine of the twenty-one bundled tools read public data and never touch the target. The other twelve — nmap, nuclei, naabu, ffuf, katana — stay refused until authorization exists for that target.
- The gate is in the executor
- A tool classified as active is refused unless the orchestrator supplies an authorization token. The check sits below the API and below the UI, so it cannot be clicked past.
- Ownership you prove yourself
- Publish a DNS TXT record on a domain you control and the platform writes the authorization record for it. No admin ticket, and no scope beyond that domain.
- Containers with nothing in them
- Read-only root filesystem, all Linux capabilities dropped, no-new-privileges, memory / CPU / PID limits, and a scoped network. Arguments are templated into argv — a shell is never involved.
- Tools hold no secrets
- API, orchestrator and tool workers are separate processes. Credentials and internal network access never reach a tool container.
- Tenant scoping and roles
- Every query is tenant-scoped at the repository layer, and organization access is role-based across owner, admin, member and viewer.
- An append-only trail
- Missions, authorizations and administrative actions are written to an audit log that is added to and never edited.
These are controls that exist in the code. Zero-Recon claims no compliance certification and holds no third-party audit — if that is a requirement for you, it is not met today.
Active tools stay refused until a signed authorization record exists for the target. Proving you control the DNS zone writes one.
- TYPE
- TXT
- NAME
- _zerorecon.acme-demo.example
- VALUE
- zr-verify=7c1f…a904
Capabilities
What the platform does today.
Five things, all of them shipping. Anything not on this list is not built yet.
Planning that reacts to evidence
plan → execute → analyze → replan
The agent builds a multi-phase task graph rather than replaying a fixed pipeline, and revises it when a result changes what is worth doing next.
One entity graph per target
deduplicated · confidence-scored
Domains, addresses, services and findings are merged into a single queryable graph with deterministic identities, so the same host discovered by three tools stays one host.
Reports bounded by coverage
md · json
Markdown and JSON artifacts carrying the findings, the graph and an explicit statement of what was not measured.
Tools are registry rows
cli · api · script
A tool is a name, an execution definition, input and output schemas, a cost model and a mandatory side-effect class. Twenty-one ship with the platform; more are added from the dashboard without a code change or a redeploy.
No single model dependency
provider-agnostic
Gemini, OpenAI and Anthropic sit behind one interface chosen by configuration. The deterministic half of the platform — coverage, scoring, the authorization gate — keeps working when a provider does not.
Personal & team
One operator, or an organization with roles.
A workspace is created as one or the other. Which one you have decides what the app lets you do — and when something is unavailable, it says why instead of failing on click.
Personal
Your own missions, reports and coin balance. Everything on this page works in a personal workspace except the collaboration column opposite.
- Missions, plans and reports scoped to you
- Passive recon on the free daily allowance
- Domain ownership verification for your own targets
Team
An organization other people belong to, with a role each and an email invitation flow to get them in.
| Role | Can |
|---|---|
| Owner | Everything, including transferring ownership |
| Admin | Invite members, change roles, remove members |
| Member | Run and read the organization's missions |
| Viewer | Read only |
A personal workspace can be converted to a team workspace from the Teams page. Until it is, team controls are shown disabled with the server's own reason attached — never enabled and then refused.
Getting started
Start free. Pay for work performed.
Usage is metered in Z-Coins: tool executions cost by runtime and resources, an ordinary chat message costs five. Run out mid-mission and the mission pauses — you keep the partial report, and the steps that never ran count against its coverage.
Free
150 coins / day
The full workspace, passive reconnaissance, missions and reports. The allowance refreshes daily.
Create an accountPremium
4,500 coins / month
Higher limits, queue priority, longer context. There is no self-serve checkout yet — Premium is arranged directly, and the pricing page has the route.
See PremiumCustom
By arrangement
Teams and organizations with their own coin allowance and limits. Describe what you need and we size it.
Request a planFAQ
Questions worth asking first.
What is Zero-Recon?
A reconnaissance platform driven by an agent. You state a target in the workspace; it plans a multi-phase mission from its tool registry, asks you to approve the plan, runs the approved tools in isolated containers, merges the output into one entity graph, and produces a report whose conclusions are bounded by how much of the plan actually completed.
Who is it for?
Security engineers, penetration testers and bug bounty hunters who already run subfinder, dnsx, httpx, nuclei and nmap by hand, and small security teams that want those runs recorded, scoped and shared rather than living in someone's terminal history.
Does it replace the tools I already use?
No — it runs them. The bundled twenty-one are the same open-source binaries and APIs you would run yourself, executed in sandboxed containers. What Zero-Recon replaces is the orchestration around them: the ordering, the correlation, the authorization check and the reporting.
What counts as active scanning?
Any tool whose registry row classifies it as having active side effects — it sends traffic to the target rather than reading public records about it. nmap, nuclei, naabu, ffuf and katana are active. subfinder, dnsx, httpx, tlsx, cdncheck, whois, wayback, crt.sh and certspotter are passive and run without authorization.
How is authorization handled?
An active tool will not execute unless the orchestrator supplies an authorization token, which exists only once a signed authorization record covers the target. You can create one yourself by proving control of the domain with a DNS TXT record; the record then unlocks active tools for that domain and nothing else. The check lives in the executor, so it applies regardless of what the API or the interface allows.
Can a team use it together?
Yes, in a team workspace: organizations with owner, admin, member and viewer roles, email invitations, member management and shared missions. A workspace created as personal has those capabilities locked until it is converted, and the app states that rather than offering controls that fail.
Can I sign up with an email address?
Email registration with verification is built, alongside Google and GitHub sign-in. Whether email sign-up is open is a deployment setting an administrator controls, so on some instances the OAuth routes are the way in.
What happens to my scan data?
It stays in your workspace. Every query is tenant-scoped at the repository layer, so missions, findings and reports are readable only within the account or organization that produced them, and tool containers receive no credentials and no internal network access. There is no third-party audit of this — it is an implementation detail you are welcome to verify.
Point it at a target you own.
Passive reconnaissance runs on the free daily allowance. Nothing active happens until you prove you are allowed to run it.