Alpha release

Stateful primitives for AI agents — MCP + REST

Your agents are stateless. Their obligations aren't.

AIGears gives AI agents the four primitives they can't provide for themselves: scheduled callbacks, deadline memory, human approval gates, and watch-and-notify. Durable infrastructure that outlives any session, context window, or process.

Start free Explore the API

Free tier. No card. First tool call in minutes.

schedule_create

// agent calls MCP tool
schedule_create({
  "fire_at": "2026-06-17T09:00:00Z",
  "webhook_url": "https://api.yourapp.com/hooks/agent",
  "payload": { "task": "renew_ssl_cert", "domain": "example.com" }
})

// AIGears responds — and remembers
{ "id": "sch_8f3ka92", "status": "pending" }

// 7 days later, your webhook fires. Signed. Retried. Logged.

Stateless agents need stateful infrastructure.

No matter how capable models get, they cannot exist across time and sessions. Agent developers hit the same four walls — and rebuild the same brittle workarounds with cron, Redis, and ad-hoc webhook handlers.

01

Agents forget.

Across sessions, context resets, and handoffs, critical commitments — "renew this on March 14" — silently disappear.

02

Agents can't wait.

There is no native way to say "do X in 3 days." An agent cannot schedule past its own process death.

03

Agents shouldn't act unilaterally.

Sending money, deleting data, contacting customers — real side effects need human sign-off, and there's no clean primitive for "ask a human, block until answered."

04

Agents can't watch.

They can fetch a URL once. They cannot monitor it for change without persistent infrastructure.

AIGears is that infrastructure — built once, properly, and shared by all four tools.

Four tools. One backend. One story.

Every tool ships with retries, HMAC-signed webhooks, idempotency keys, and an append-only audit log — because they all sit on the same core.

schedule_create

Scheduled callbacks

Register a future webhook. We fire it.

One-shot or recurring with cron expressions. Timezone-aware. Survives everything your agent doesn't.

→Retries with exponential backoff on failure
→Idempotency keys prevent duplicate registration
→Full delivery history per callback

POST /v1/schedules/

curl -X POST https://aigears.net/v1/schedules/ \
  -H "Authorization: Bearer $AIGEARS_API_KEY" \
  -d '{
    "recurrence": "0 9 * * MON",
    "recurrence_tz": "Europe/Paris",
    "webhook_url": "https://api.yourapp.com/hooks/weekly-report",
    "payload": { "job": "compile_weekly_report" }
  }'

deadline_remember

Deadline memory

Durable, queryable memory for dated obligations.

A new agent instance can ask "what did past-self commit to?" and get an answer. Escalation policies notify the agent by webhook, a human by email, or both — at 30 days, 7 days, day-of, overdue.

→Survives sessions, resets, and handoffs
→Query by window and tags
→Arbitrary JSON metadata — your schema, not ours

deadline_remember

deadline_remember({
  "due_at": "2027-03-14T00:00:00Z",
  "summary": "Renew ACME contract before auto-expiry",
  "tags": ["contracts", "compliance"],
  "escalation_policy": ["30d", "7d", "1d", "overdue_daily"],
  "notify_emails": ["[email protected]"],
  "webhook_url": "https://api.yourapp.com/hooks/deadlines"
})

approval_request

Approval gates

Coming soon

Block agent action on human sign-off.

The missing primitive in every "agent with guardrails" architecture. We deliver a signed, single-use URL to the approver; the hosted page shows full action context before any button is clickable. One approval can guard a $10,000 action.

→Approve or reject with a comment, callback or polling
→Timeouts produce an explicit expired status
→Audit trail: who approved, when, from what IP

approval_request

approval_request({
  "action_summary": "Refund customer #4821 — $1,250.00",
  "action_details": { "order": "ord_55310", "reason": "duplicate charge" },
  "approver_channel": "email:[email protected]",
  "timeout": "24h",
  "callback_url": "https://api.yourapp.com/hooks/approvals"
})
// → { "approval_id": "apr_2c9x", "status": "pending" }

watch_create

Watch-and-notify

Coming soon

Register a source. We diff it. You get a webhook on change.

Converts pull-once into push-on-change. URL, RSS, or JSON endpoints, with CSS-selector extraction and match rules.

→Webhook fires with old and new values
→Equality, regex, and JSONPath match rules
→Respects robots.txt and Cache-Control

watch_create

watch_create({
  "source_type": "json_endpoint",
  "source_config": {
    "url": "https://status.vendor.com/api/v2/status.json",
    "jsonpath": "$.status.indicator"
  },
  "match_rule": { "changed": true },
  "webhook_url": "https://api.yourapp.com/hooks/vendor-status",
  "poll_interval": "5m"
})

One backend. Two interfaces.

Agents consume AIGears natively over MCP with zero glue code. Everything is equally available as plain REST for LangChain, custom Python, or anything with an HTTP client.

MCP

One hosted MCP server, all four toolsets. Strict JSON schemas on every tool — agents perform best with tight contracts.

schedule_create · schedule_cancel · schedule_list · schedule_get · deadline_remember · deadline_query · deadline_complete · deadline_snooze · approval_request · approval_status · approval_cancel · watch_create · watch_pause · watch_history …

REST

Versioned /v1/ API. JSON in, JSON out, conventional HTTP semantics, and a published OpenAPI spec.

Browse the OpenAPI spec →

Every webhook, signed and accounted for.

HMAC-SHA256, Stripe-style: Timestamped signatures with a replay window and secret rotation grace period.
Retries with backoff: Exponential backoff on 5xx and timeouts, with a dead-letter queue. Nothing is silently dropped.
Append-only audit log: Every tool call, every webhook fired, every approval decision. Exportable.

incoming webhook

POST /hooks/agent HTTP/1.1
X-AIGears-Signature: t=1750147200,v1=5257a86...
Content-Type: application/json

{ "event": "schedule.fired", "schedule_id": "sch_8f3ka92" }

See everything your agents have done.

The dashboard isn't where the work happens — your agents do that. It's how you trust the work: a chronological activity feed, upcoming callbacks and deadlines, pending approvals, active watches, API keys, and a filterable audit log.

aigears.net/a/acme/activity

Time	Event	Detail	Result
09:00:01	schedule.fired	sch_8f3ka92 → /hooks/agent	✓ 200
08:42:17	approval.decided	apr_2c9x — approved by [email protected]	✓
08:30:55	watch.changed	vendor-status: operational → degraded	↻ fired
07:15:09	deadline.escalation	ACME contract — 7d notice	email sent

Read-heavy by design. Most state is created by the agent, not the human.

Pricing that scales with state, not calls.

Persistent state and polling cost us; individual API calls don't. Pricing follows the same shape. Start on the free tier — 50 scheduled callbacks, 25 active deadlines, 10 approvals, and 5 watches per month — and upgrade when your agents outgrow it.

Start free See full pricing →