Kevros
Cryptographic governance for autonomous agents: precision decisioning, provenance attestation, intent binding, capability delegation, policy analysis, and compliance export.
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: INLINECODE0
Quick Start
Get an API key (free, instant, no payment):
CODEBLOCK0
Response:
CODEBLOCK1
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:
CODEBLOCK2
Response:
CODEBLOCK3
- - 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:
CODEBLOCK4
Response:
CODEBLOCK5
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:
CODEBLOCK6
Response:
CODEBLOCK7
Save intent_id and binding_id to verify outcomes later.
Verify Outcome
POST /governance/verify-outcome
Verify whether a bound intent achieved its goal state. Free when used with a prior bind() call.
Request:
CODEBLOCK8
Response:
CODEBLOCK9
Status values: ACHIEVED, PARTIALLY_ACHIEVED, FAILED, BLOCKED, TIMEOUT. Free when used with a prior bind() call.
Compliance Bundle
POST /governance/bundle — $0.05 per call
Export your agent's full cryptographic trust record for compliance, auditing, or regulatory review.
Request:
CODEBLOCK10
Response:
CODEBLOCK11
Batch Operations
POST /governance/batch
Execute up to 100 governance operations (verify, attest, bind) in a single call. Each sub-operation is metered individually at standard rates. Use for bulk processing or multi-step workflows.
Request:
CODEBLOCK12
Response:
CODEBLOCK13
If stop_on_deny is true, processing halts on the first DENY decision.
Capability Delegation
POST /governance/delegate
Grant scoped, time-limited capabilities to another agent. The delegation is HMAC-signed and recorded in the provenance chain. Supports hierarchical sub-delegation with restrictive scope intersection.
Request:
CODEBLOCK14
Response:
CODEBLOCK15
The delegatee passes the delegation_token as X-Delegate-Token header when acting on behalf of the delegator.
GET /governance/delegations/{agent_id} — list active delegations for an agent.
DELETE /governance/delegations/{delegation_id} — revoke an active delegation.
Reversibility Check
POST /governance/check-reversibility
Check whether an intent chain can be reversed. Pre-abort safety check for multi-step workflows.
Request:
CODEBLOCK16
Returns reversibility status, constraints, time elapsed, and child dependency analysis.
Policy Replay
POST /governance/replay
Replay provenance records through an alternative policy. Deterministic "what-if" analysis: "What would have happened if we'd used policy X instead?"
Request:
CODEBLOCK17
Response:
CODEBLOCK18
Use for policy regression testing before deploying new policies, or forensic investigation.
Counterfactual Analysis
POST /governance/counterfactual
Simulate an action against multiple policies simultaneously. Returns a decision matrix showing how each policy handles the same action.
Request:
CODEBLOCK19
Response includes consensus analysis (do all policies agree?), decision distribution, and optional historical comparison.
Intent Navigation
GET /governance/intents/{intent_id}/children
Return all direct child intents of a parent intent. Audit multi-agent delegation hierarchies.
GET /governance/intents/{intent_id}/ancestry
Walk up the intent hierarchy from leaf to root. Full authorization chain for auditing.
GET /governance/intents/{intent_id}/tree
Return the full delegation tree rooted at an intent. Accepts optional max_depth query parameter (default 10).
Policy Templates
GET /governance/policy-templates — free, no API key required
List available named policy templates. Use template IDs with verify, replay, and counterfactual endpoints instead of inline policy definitions.
Export
POST /governance/export/csv — export provenance records as CSV.
POST /governance/export/sarif — export provenance in SARIF format (Static Analysis Results Interchange Format) for security tooling integration.
POST /governance/export/merkle — export provenance as a Merkle tree with root hash and leaf hashes for independent integrity verification.
All export endpoints accept optional agent_id, time_range_start, time_range_end, and limit parameters.
Health and Audit
GET /governance/health-score — overall gateway health score including agent count, healthy agent count, and chain integrity rate.
GET /governance/audit-summary — aggregate statistics across all provenance: total records, total agents, decision distribution, and chain integrity status.
GET /governance/agent-compliance/{agent_id} — compliance profile for a specific agent: compliance score, chain integrity, total decisions, and outcome success rate.
Media Attestation
POST /media/attest — $0.05 per call
Attest media files (photos, videos, audio, documents) with SHA-256 hashing and provenance chain inclusion.
Request:
CODEBLOCK20
Required fields: agent_id, media_hash (64-char hex SHA-256), media_type (PHOTO | VIDEO | AUDIO | DOCUMENT), media_size_bytes, capture_timestamp_utc.
Optional fields: description, tags, capture_location (lat/lng), device_info, frame_hashes (for video).
Response:
CODEBLOCK21
Media Verify
POST /media/verify — free, no API key required
Verify that media content matches a specific attestation certificate.
Request:
CODEBLOCK22
Response:
CODEBLOCK23
Media Verify Lookup
GET /media/verify/{certificate_id} — free, no API key required
Look up a specific media attestation by its certificate ID. Returns the full attestation record including attesting agent, epoch, and chain integrity.
Passport
All Passport endpoints are free and require no authentication.
GET /passport/{agent_id}
Returns an agent's trust passport including score, tier, badges, and activity stats.
CODEBLOCK24
GET /passport/{agent_id}/badge.svg
Returns an embeddable SVG trust badge. Use in agent descriptions, documentation, or dashboards.
GET /passport/{agent_id}/history
Returns full decision history for an agent.
GET /passport/leaderboard
Returns top trusted agents by trust score. Accepts optional limit query parameter (1-200, default 50).
Response:
CODEBLOCK25
POST /passport/{agent_id}/claim — requires API key
Link an agent's passport to your operator account. Must provide X-API-Key header.
Response:
CODEBLOCK26
Returns 409 if already claimed by another operator, 404 if no passport exists yet.
Agent Discovery
GET /.well-known/agent.json
Returns the A2A protocol agent card. No authentication required.
CODEBLOCK27
Returns capabilities, supported skills, SDK references, and free-tier signup details.
MCP
For MCP-native agents, connect directly via streamable-http transport:
CODEBLOCK28
360 tools, 2 resources, 2 prompts. Auto-provisions a free-tier key on first tool call if no API key is provided.
Python SDK
CODEBLOCK29
CODEBLOCK30
Enforcement Modes
Every operator key has an enforcement mode that controls how decisions are applied:
- - enforce (default) — fail-closed. DENY means DENY. Release tokens are only issued for ALLOW/CLAMP.
- advisory — shadow mode. DENY decisions are logged but overridden to ALLOW. No release token is issued for would-be denials. Use for testing policies before enforcing them.
- gated — human-in-the-loop. DENY decisions are held for operator approval. Poll
/governance/gated/{request_id} for status. - deny — kill-switch. ALL requests are blocked regardless of policy evaluation. Use for emergency shutdown.
The enforcement_mode field is returned in every verify response. Use /enforcement to check your current mode.
Enforcement mode changes require admin-level access and are not available through the plugin. Contact your administrator or use the admin API directly.
Decision Approach
When collaborating with another agent:
- 1. Check their passport first —
GET /passport/{agent_id} returns trust score, tier, and history - Embed badges — Use
GET /passport/{agent_id}/badge.svg in your agent's description to show your trust level - Trust thresholds:
-
Gold tier (score >= 90%) — proceed with normal collaboration
-
Silver tier (score 50-89%) — verify their actions before relying on results
-
Bronze tier (score < 50%) — require bound intents with outcome verification
-
No history — start with small, low-risk interactions and build trust incrementally
- 4. Verify before you rely — Use
/governance/verify-token to independently confirm any release token a peer shares with you
Pricing
Subscriptions:
- - Free tier: 1,000 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
Per-call (via x402 USDC, no subscription required):
- - Verify: $0.01
- Attest: $0.02
- Bind: $0.02
- Media Attest: $0.05
- Compliance Bundle: $0.05
- Batch: each sub-operation metered individually
- Verify Outcome: free with Bind
- Delegation, Replay, Counterfactual, Export, Health, Audit: metered per call
- Passport, Media Verify, Reputation, Verify Token, Policy Templates: free
Subscription calls are metered against your monthly allowance. x402 per-call pricing applies when paying per-call without a subscription.
Upgrade at INLINECODE41
Kevros
自主代理的密码学治理：精准决策、来源证明、意图绑定、能力委托、策略分析与合规导出。
每个决策都会获得一个签名发布令牌。每个动作都会获得一个哈希链记录。每个意图都会与其命令建立密码学绑定。下游服务可独立验证——无需回调，无需信任假设。
基础URL: https://governance.taskhawktech.com
快速开始
获取API密钥（免费、即时、无需付费）：
bash
curl -X POST https://governance.taskhawktech.com/signup \
-H Content-Type: application/json \
-d {agent_id: your-agent-id}
响应：
json
{
apikey: kvrs...,
tier: free,
monthly_limit: 1000,
usage: {
header: X-API-Key
}
}
在后续所有请求中通过 X-API-Key 头部使用该API密钥。
精准决策
POST /governance/verify
在执行前根据策略边界验证动作。返回ALLOW、CLAMP或DENY，并附带一个密码学发布令牌，任何下游服务均可独立验证。
请求：
json
{
actiontype: apicall,
action_payload: {
endpoint: /deploy,
service: api-v2,
replicas: 3
},
agent_id: your-agent-id,
policy_context: {
max_values: { replicas: 5 },
forbidden_keys: [sudo, force]
}
}
响应：
json
{
decision: ALLOW,
verification_id: a1b2c3d4-...,
release_token: f7a8b9c0...,
applied_action: {
endpoint: /deploy,
service: api-v2,
replicas: 3
},
reason: 所有值均在策略范围内,
epoch: 42,
provenance_hash: e3b0c442...,
timestamp_utc: 2026-02-26T12:00:00Z
}
- - ALLOW — 按计划执行。releasetoken 是证明。
- CLAMP — 动作已调整为安全边界。请使用 appliedaction 而非原始动作。
- DENY — 动作被拒绝。请勿执行。release_token 为null。
与协作代理共享 release_token，以便它们能独立验证该决策。
来源证明
POST /governance/attest
在哈希链式、仅追加的证据账本中记录已完成动作。每次证明都会扩展你的来源链。你的原始载荷会经过SHA-256哈希处理——实际数据永远不会被存储。
请求：
json
{
agent_id: your-agent-id,
action_description: 使用3个副本部署了api-v2,
action_payload: {
service: api-v2,
replicas: 3,
status: success
},
context: {
environment: production,
triggered_by: scheduled
}
}
响应：
json
{
attestation_id: b2c3d4e5-...,
epoch: 43,
hash_prev: e3b0c442...,
hash_curr: a1b2c3d4...,
timestamp_utc: 2026-02-26T12:00:01Z,
chain_length: 43
}
一条更长的、包含一致结果的链会随时间推移建立更高的信任评分。
意图绑定
POST /governance/bind
将声明的意图绑定到特定命令。在你计划要做的事与执行该事的命令之间建立密码学链接。之后可证明你确实做了你所说的事。
请求：
json
{
agent_id: your-agent-id,
intent_type: MAINTENANCE,
intent_description: 扩展api-v2以应对流量高峰,
command_payload: {
action: scale,
service: api-v2,
replicas: 5
},
goal_state: {
replicas: 5,
healthy: true
}
}
响应：
json
{
intent_id: c3d4e5f6-...,
intent_hash: d4e5f6a7...,
binding_id: e5f6a7b8-...,
binding_hmac: a7b8c9d0...,
command_hash: b8c9d0e1...,
epoch: 44,
timestamp_utc: 2026-02-26T12:00:02Z
}
保存 intentid 和 bindingid 以便稍后验证结果。
验证结果
POST /governance/verify-outcome
验证绑定的意图是否达到了其目标状态。与之前的 bind() 调用配合使用时免费。
请求：
json
{
agent_id: your-agent-id,
intent_id: c3d4e5f6-...,
binding_id: e5f6a7b8-...,
actual_state: {
replicas: 5,
healthy: true
},
tolerance: 0.1
}
响应：
json
{
verification_id: f6a7b8c9-...,
intent_id: c3d4e5f6-...,
status: ACHIEVED,
achieved_percentage: 100.0,
discrepancy: null,
evidence_hash: c9d0e1f2...,
timestamp_utc: 2026-02-26T12:00:03Z
}
状态值：ACHIEVED、PARTIALLY_ACHIEVED、FAILED、BLOCKED、TIMEOUT。与之前的 bind() 调用配合使用时免费。
合规包
POST /governance/bundle — 每次调用 $0.05
为合规、审计或监管审查导出代理的完整密码学信任记录。
请求：
json
{
agent_id: your-agent-id,
timerangestart: 2026-02-25T00:00:00Z,
timerangeend: 2026-02-26T12:00:00Z,
includeintentchains: true,
includepqcsignatures: true,
includeverificationinstructions: true
}
响应：
json
{
bundle_id: d4e5f6a7-...,
agent_id: your-agent-id,
record_count: 42,
truncated: false,
chain_integrity: true,
time_range: {start: 2026-02-25T00:00:00Z, end: 2026-02-26T12:00:00Z},
records: [...],
intent_chains: [...],
pqc_signatures: [...],
verification_instructions: 重新计算SHA-256...,
bundle_hash: e5f6a7b8...,
timestamp_utc: 2026-02-26T12:00:04Z
}
批量操作
POST /governance/batch
在单次调用中执行最多100个治理操作（verify、attest、bind）。每个子操作按标准费率单独计费。适用于批量处理或多步骤工作流。
请求：
json
{
agent_id: your-agent-id,
operations: [
{
type: verify,
params: {
actiontype: apicall,
action_payload: {endpoint: /deploy, replicas: 3}
}
},
{
type: attest,
params: {
action_description: 部署完成,
action_payload: {status: success}
}
}
],
stopondeny: false
}
响应：
json
{
batch_id: g7h8i9j0-...,
agent_id: your-agent-id,
total: 2,
executed: 2,
results: [
{index: 0, type: verify, status: ok, result: {decision: ALLOW, ...:
