GET /health

Returns the current status of the SENTINEL API. No authentication required.

Example Request

curl https://api.starksentinel.com/health

Example Response

{
  "status": "ok",
  "version": "1.3.1",
  "service": "sentinel-attestation-service"
}

Response Fields

POST /v1/prove

Generates a STARK proof attesting that a set of behavioral observations stays within defined EWMA control limits. The proof is zero-knowledge: raw observations are used to compute the EWMA in-memory, then discarded. Only public inputs (observation count, final EWMA, control limits, pass/fail) are included in the proof envelope.

Request Body

Example Request

# Generate a STARK proof for 8 behavioral observations
curl -X POST https://api.starksentinel.com/v1/prove \
  -H "X-API-Key: $SENTINEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "observations": [95, 98, 92, 97, 100, 94, 96, 99],
    "baseline_mean": 96,
    "ucl": 110,
    "lcl": 80
  }'

Example Response

{
  "version": "1.2.0",
  "proof_bytes": [4, 0, 0, 4, ...],
  "public_inputs": {
    "num_observations": 8,
    "first_observation": 95,
    "final_ewma": 95,
    "lambda_num": 1,
    "lambda_den": 5,
    "baseline_mean": 96,
    "ucl": 110,
    "lcl": 80,
    "within_limits": true
  },
  "metadata": {
    "trace_length": 16,
    "num_constraints": 3,
    "security_level": 97,
    "generation_time_ms": 2,
    "proof_size_bytes": 5995
  },
  "chain_hash": "72f2dd09...8b0c33"
}

Response Fields

Privacy Guarantee

Individual observation values are never included in the response. The proof attests to the statistical properties (EWMA within control limits) without revealing the underlying data. This is the Prove-Don't-Share architecture.

POST /v1/verify

Verifies a previously generated STARK proof. Verification confirms that the proof was correctly generated and that the public inputs are authentic. Verification is always free and does not count against your monthly proof limit.

Request Body

Send the complete proof envelope — the entire JSON response from /v1/prove. All fields are required.

Example Request

# Save the prove response, then pass it directly to verify
curl -s -X POST https://api.starksentinel.com/v1/prove \
  -H "X-API-Key: $SENTINEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"observations": [95,98,92,97,100,94,96,99], "baseline_mean": 96, "ucl": 110, "lcl": 80}' > proof.json

# Verify the full proof envelope
curl -X POST https://api.starksentinel.com/v1/verify \
  -H "X-API-Key: $SENTINEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d @proof.json

Example Response

{
  "valid": true
}

Response Fields

Client-Side Verification

For zero-trust verification, use the open-source WASM verifier. It runs entirely client-side — no network calls, no trust in SENTINEL required. See the Quickstart Guide for setup instructions.

POST /v1/verify-chain

Verifies a chain of linked STARK proofs. Checks that each proof is cryptographically valid and that chain links are intact — each proof's previous_proof_hash matches the preceding proof's chain_hash. The first proof in the chain must have a null previous hash (genesis proof).

Request Body

A JSON array of proof envelopes, ordered from genesis (first) to most recent (last). Maximum 100 proofs per request.

Example Request

# Verify a 2-proof chain
curl -X POST https://api.starksentinel.com/v1/verify-chain \
  -H "X-API-Key: $SENTINEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    { "version": "1.2.0", "proof_bytes": [...], "public_inputs": {...},
      "chain_hash": "a1b2c3...", "previous_proof_hash": null, "metadata": {...} },
    { "version": "1.2.0", "proof_bytes": [...], "public_inputs": {...},
      "chain_hash": "d4e5f6...", "previous_proof_hash": "a1b2c3...", "metadata": {...} }
  ]'

Example Response

{
  "valid": true,
  "chain_length": 2,
  "valid_proofs": 2,
  "valid_links": 2,
  "failed_proofs": [],
  "broken_links": []
}

Response Fields

POST /v1/aggregate

Aggregates a chain of proofs into a single Merkle commitment. Verifies all proofs and chain links, then produces a compact aggregate with a Merkle root, summary statistics, and per-proof summaries. Maximum 10,000 proofs per request.

Request Body

Same format as /v1/verify-chain — a JSON array of proof envelopes ordered from genesis to most recent.

Example Response

{
  "version": "1.0.0",
  "chain_id": "my-agent-chain",
  "chain_length": 2,
  "merkle_root": "f8a3b1c9d2e4...",
  "aggregate_inputs": {
    "first_observation": 95,
    "final_ewma": 97,
    "all_within_limits": true,
    "min_security_level": 128,
    "total_observations": 16
  },
  "proof_summaries": [
    {
      "chain_hash": "a1b2c3...",
      "within_limits": true,
      "num_observations": 8,
      "final_ewma": 95,
      "security_level": 128
    }
  ],
  "metadata": {
    "aggregation_time_ms": 1,
    "proofs_verified": 2,
    "links_verified": 1,
    "aggregated_at": "2026-04-03T12:00:00.000Z"
  }
}

Response Fields

POST /v1/verify-aggregate

Verifies a previously produced aggregate by recomputing the Merkle root from the proof summaries and confirming it matches. This is a lightweight check — no STARK proofs need to be re-verified.

Request Body

Pass the full aggregate response object from /v1/aggregate.

Example Response

{
  "valid": true
}

Response Fields

The Proof Registry is a separate service at registry.starksentinel.com. It provides immutable, publicly queryable storage for proof records.

Registry Endpoints

Base URL

https://registry.starksentinel.com

Register a Proof

# Register a proof after generating and verifying it
curl -X POST https://registry.starksentinel.com/v1/proofs \
  -H "X-API-Key: $SENTINEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "chain_hash": "a1b2c3d4e5f6...",
    "previous_proof_hash": null,
    "version": "1.2.0",
    "publisher": "my-company",
    "chain_id": "agent-quality-chain",
    "num_observations": 8,
    "final_ewma": 95,
    "within_limits": true,
    "ucl": 110,
    "lcl": 80,
    "baseline_mean": 96,
    "security_level": 128,
    "proof_size_bytes": 6761
  }'

Look Up a Proof

# Anyone can look up a registered proof — no API key needed
curl https://registry.starksentinel.com/v1/proofs/a1b2c3d4e5f6...

Example Lookup Response

{
  "id": 1,
  "chain_hash": "a1b2c3d4e5f6...",
  "previous_proof_hash": null,
  "version": "1.2.0",
  "publisher": "my-company",
  "chain_id": "agent-quality-chain",
  "num_observations": 8,
  "final_ewma": 95,
  "within_limits": true,
  "ucl": 110,
  "lcl": 80,
  "baseline_mean": 96,
  "security_level": 128,
  "proof_size_bytes": 6761,
  "registered_at": "2026-04-03T12:00:00.000Z"
}

Registration is immutable — once a proof is registered, it cannot be modified or deleted. Attempting to register a duplicate chain_hash returns 409 Conflict.

MCP sentinel_compute_bid

Calculates a Behavioral Injection Score (BIS) locally by comparing 5 behavioral metrics against their baselines. Returns a composite 0-100 score, a response level, and per-metric sigma distances. Runs entirely in-process — no API call required.

Parameters

Example Request

# Via MCP tool call
{
  "enforcementRate": 0.35,
  "enforcementRate_mean": 0.05,
  "enforcementRate_stdDev": 0.02,
  "targetEntropy": 0.8,
  "targetEntropy_mean": 2.1,
  "targetEntropy_stdDev": 0.3,
  "actionFrequency": 45,
  "actionFrequency_mean": 12,
  "actionFrequency_stdDev": 3,
  "typeConcentration": 0.9,
  "typeConcentration_mean": 0.4,
  "typeConcentration_stdDev": 0.1,
  "latencyMs": 15,
  "latencyMs_mean": 5,
  "latencyMs_stdDev": 2
}

Example Response

{
  "bis": 82.4,
  "response_level": "PAUSE",
  "metrics": {
    "enforcementRate": { "value": 0.35, "sigma": 15.0, "weight": 0.35 },
    "targetEntropy": { "value": 0.8, "sigma": -4.33, "weight": 0.25 },
    "actionFrequency": { "value": 45, "sigma": 11.0, "weight": 0.20 },
    "typeConcentration": { "value": 0.9, "sigma": 5.0, "weight": 0.15 },
    "latencyMs": { "value": 15, "sigma": 5.0, "weight": 0.05 }
  }
}

Response Levels

MCP sentinel_sanitize_input

Strips invisible Unicode characters, reasoning tags, and model control tokens from input text. Returns the cleaned text with a breakdown of what was removed. Runs locally — no API call required.

Parameters

Example Response

{
  "cleaned": "Please summarize the quarterly report",
  "strippedCount": 14,
  "categories": {
    "tags_block": 0,
    "zero_width": 8,
    "bidi_control": 2,
    "reasoning_tags": 1,
    "model_control": 3
  }
}

What Gets Stripped