Kevros

Kevros

Three cryptographic primitives for autonomous agents: precision decisioning, provenance attestation, and intent binding.

Every decision gets a signed release token. Every action gets a hash-chained record. Every intent gets a cryptographic binding to its command. Downstream services verify independently — no callbacks, no trust assumptions.

Base URL: https://governance.taskhawktech.com

Quick Start

Get an API key (free, instant, no payment):

curl -X POST https://governance.taskhawktech.com/signup \
  -H "Content-Type: application/json" \
  -d '{"agent_id": "your-agent-id"}'

Response:

{
  "api_key": "kvrs_...",
  "tier": "free",
  "monthly_limit": 100,
  "usage": {
    "header": "X-API-Key"
  }
}

Use the API key in all subsequent requests via the X-API-Key header.

Precision Decisioning

POST /governance/verify

Verify an action against policy bounds before execution. Returns ALLOW, CLAMP, or DENY with a cryptographic release token that any downstream service can verify independently.

Request:

{
  "action_type": "api_call",
  "action_payload": {
    "endpoint": "/deploy",
    "service": "api-v2",
    "replicas": 3
  },
  "agent_id": "your-agent-id",
  "policy_context": {
    "max_values": { "replicas": 5 },
    "forbidden_keys": ["sudo", "force"]
  }
}

Response:

{
  "decision": "ALLOW",
  "verification_id": "a1b2c3d4-...",
  "release_token": "f7a8b9c0...",
  "applied_action": {
    "endpoint": "/deploy",
    "service": "api-v2",
    "replicas": 3
  },
  "reason": "All values within policy bounds",
  "epoch": 42,
  "provenance_hash": "e3b0c442...",
  "timestamp_utc": "2026-02-26T12:00:00Z"
}

• ALLOW — proceed as planned. The release_token is proof.
• CLAMP — action was adjusted to safe bounds. Use applied_action instead of your original.
• DENY — action rejected. Do not proceed. release_token is null.

Share the release_token with collaborating agents so they can independently verify the decision.

Provenance Attestation

POST /governance/attest

Record a completed action in a hash-chained, append-only evidence ledger. Each attestation extends your provenance chain. Your raw payload is SHA-256 hashed — actual data is never stored.

Request:

{
  "agent_id": "your-agent-id",
  "action_description": "Deployed api-v2 with 3 replicas",
  "action_payload": {
    "service": "api-v2",
    "replicas": 3,
    "status": "success"
  },
  "context": {
    "environment": "production",
    "triggered_by": "scheduled"
  }
}

Response:

{
  "attestation_id": "b2c3d4e5-...",
  "epoch": 43,
  "hash_prev": "e3b0c442...",
  "hash_curr": "a1b2c3d4...",
  "timestamp_utc": "2026-02-26T12:00:01Z",
  "chain_length": 43
}

A longer chain with consistent outcomes builds a higher trust score over time.

Intent Binding

POST /governance/bind

Bind a declared intent to a specific command. Creates a cryptographic link between what you plan to do and the command that does it. Prove later that you did exactly what you said you would.

Request:

{
  "agent_id": "your-agent-id",
  "intent_type": "MAINTENANCE",
  "intent_description": "Scale api-v2 to handle traffic spike",
  "command_payload": {
    "action": "scale",
    "service": "api-v2",
    "replicas": 5
  },
  "goal_state": {
    "replicas": 5,
    "healthy": true
  }
}

Response:

{
  "intent_id": "c3d4e5f6-...",
  "intent_hash": "d4e5f6a7...",
  "binding_id": "e5f6a7b8-...",
  "binding_hmac": "a7b8c9d0...",
  "command_hash": "b8c9d0e1...",
  "epoch": 44,
  "timestamp_utc": "2026-02-26T12:00:02Z"
}

Save intent_id and binding_id to verify outcomes later.

Agent Discovery

GET /.well-known/agent.json

Returns the A2A protocol agent card. No authentication required.

curl https://governance.taskhawktech.com/.well-known/agent.json

Returns capabilities, supported skills, SDK references, and free-tier signup details.

MCP

For MCP-native agents, connect directly via streamable-http transport:

https://governance.taskhawktech.com/mcp/

9 tools, 2 resources, 2 prompts. Auto-provisions a free-tier key on first tool call if no API key is provided.

Python SDK

pip install kevros
# or
uv pip install kevros

from kevros import GovernanceClient

client = GovernanceClient(agent_id="your-agent-id")
result = client.verify(action_type="api_call", action_payload={"endpoint": "/deploy"})
print(result.decision)  # ALLOW, CLAMP, or DENY

Decision Approach

When collaborating with another agent:

1. Trust score > 70% — proceed with normal collaboration
2. Trust score 40-70% — verify their actions before relying on results
3. Trust score < 40% — require bound intents with outcome verification
4. No history — start with small, low-risk interactions

Pricing

• Free tier: 100 calls/month, instant signup, no payment required
• Scout: $29/mo — 5,000 calls
• Sentinel: $149/mo — 50,000 calls
• Sovereign: $499/mo — 500,000 calls

Upgrade at https://governance.taskhawktech.com/upgrade