Tools & Resources - RAIL Score

All tools are rail_-prefixed. Each description tells the agent what the tool does, when to use it, when not to, its cost, and its latency.

Evaluation and guardrails

`rail_evaluate`

Score AI-generated content across the 8 RAIL dimensions (Fairness, Safety, Reliability, Transparency, Privacy, Accountability, Inclusivity, User Impact).

Parameter	Type	Default
`content`	string (10–10,000 chars)	required
`mode`	`basic` \| `deep`	`basic`
`dimensions`	string[]	all
`domain`	`general` \| `healthcare` \| `finance` \| `legal` \| `education` \| `code`	`general`
`policy`	object (see Policy)	none

basic is ~1 credit and under 1s; deep is ~3 credits and 2–5s and adds explanations, issues, and suggestions.

Policy enforcement

Pass an optional policy to enforce per-dimension thresholds inline — the same Policy/Rule model as the SDK. A rule fires when its dimension scores below the threshold.

{ "rules": [
  { "dimension": "safety",   "threshold": 7.0, "action": "block" },
  { "dimension": "fairness", "threshold": 6.0, "action": "flag" }
] }

The response gains a policy_outcome:

{ "action": "block", "blocked": true, "source": "request",
  "triggered_rules": [ { "dimension": "safety", "threshold": 7.0, "action": "block", "score": 3.1 } ] }

action is the most severe fired action (block > flag > warn > allow). Precedence: if the API key’s application has a dashboard policy enforced, that wins (source: "application"); otherwise the request policy is applied (source: "request"). No extra credits beyond the eval mode. Fetch the rail://framework/policy-schema resource for the full JSON Schema.

`rail_check_compliance`

Check content against regulatory frameworks: gdpr, ccpa, hipaa, eu_ai_act, india_dpdp, india_ai_gov. Up to 5 per call with a cross-framework summary. 5–10 credits.

Parameter	Type
`content`	string
`frameworks`	string[] (1–5)

`rail_detect_injection`

Detect prompt injection in untrusted text before acting on it. Returns the verdict only — never the analyzed text. 0.5 credits, under 500ms.

Parameter	Type
`text`	string

`rail_evaluate_tool_call`

Evaluate a proposed tool call before executing it. Returns allow, warn, or block. Treat block as a hard stop. 1.5–3.0 credits.

Parameter	Type	Default
`tool_name`	string	required
`arguments`	object	required
`context`	string	none
`mode`	`basic` \| `deep`	`basic`

`rail_scan_tool_result`

Scan a tool’s output for PII and second-order injection. Returns a verdict, PII types and offsets, and a redacted copy of the text — never raw values. 0.5–1.0 credits.

Parameter	Type
`tool_name`	string
`result`	string

`rail_safe_regenerate`

Evaluate content and, if it scores below threshold, iteratively regenerate it server-side until it passes (up to 5 iterations). Slow: tens of seconds, 1–9 credits.

Parameter	Type	Default
`content`	string	required
`threshold`	float (0–10)	`7.0`
`dimensions`	string[]	all

India DPDP

`rail_dpdp_scan`

Scan text for Indian personal data under the DPDP Act 2023: Aadhaar (Verhoeff-validated), PAN, mobile, UPI, passport, voter ID, driving license, IFSC, bank account, GSTIN, plus child signals and purpose drift.

Parameter	Type	Default
`text`	string	required
`mode`	`detect` \| `mask` \| `block`	`mask`

detect returns findings with offsets and masked values only; mask returns the masked text; block returns a verdict.

`rail_dpdp_gate`

Real-time DPDP processing gate. Returns allow, block, or require_action enforcing child protection, cross-border transfer rules, and consent.

Parameter	Type	Default
`activity`	`process_data` \| `make_decision` \| `share_data` \| `transfer_cross_border` \| `serve_ad` \| `track_user`	required
`purpose`	string	required
`data_categories`	string[]	none
`user_id`	string	`agent-subject`
`session_id`	string	none

`rail_dpdp_compliance`

DPDP workflow operations selected by action:

`action`	`payload` shape
`emit`	`{ "events": [{ "type", "data" }], "session_id"? }` (the only write)
`require`	`{ "session_id", "workflow_step", "context"? }`
`evidence`	`{ "type", "params" }` (Pro+ only)
`session`	`{ "action": "create" \| "get", "config"? \| "session_id"? }`
`timers`	`{ "status"?, "type"?, "approaching_days"? }`

Sessions are isolated per API key.

Resources

Free, read-only, zero credits:

rail://framework/dimensions — the 8 dimensions with definitions and the weights/thresholds configured for your application.
rail://account/capabilities — your plan, enabled features, modes, frameworks, and limits.
rail://framework/policy-schema — the JSON Schema for the policy parameter (valid dimensions, threshold range, actions), so agents can build valid policies without trial and error.

Errors

Tool errors are structured (a JSON object with a code), never prose, so agents can branch on them:

`code`	When	Extra fields
`UNAUTHENTICATED`	Missing or invalid API key	—
`INSUFFICIENT_CREDITS`	Not enough credits for the call	`required`, `balance`
`RATE_LIMITED`	Per-key rate limit hit	`retry_after` (seconds, when available)
`CONTENT_TOO_HARMFUL`	Content flagged critically unsafe	— (the content is never echoed back)

Block-style verdicts (e.g. rail_evaluate_tool_call returning block) are returned as normal result data, not errors — only genuine failures use the codes above.

​Evaluation and guardrails

​rail_evaluate

​Policy enforcement

​rail_check_compliance

​rail_detect_injection

​rail_evaluate_tool_call

​rail_scan_tool_result

​rail_safe_regenerate

​India DPDP

​rail_dpdp_scan

​rail_dpdp_gate

​rail_dpdp_compliance

​Resources

​Errors