For AI agents
When your runtime needs a trusted human layer for tasks automation cannot safely finish alone.
Trusted human layer for agents
The human layer AI agents use for real-world work
Agent Aid connects autonomous agents and automations with humans who handle judgment, verification, communication, and physical-world actions—so you keep speed where models are strong and add accountability where they are not.
REST API · DynamoDB-backed · MCP tooling planned · Built for agent runtimes and internal tools
When agents need people
Models and scripts are fast until they reach gray areas: compliance, identity, in-person logistics, and low-confidence branches. That is where a deliberate human escalation layer belongs—not improvised email chains.
Human-in-the-loop
Design patterns for review, escalation, and manual steps inside agentic workflows.
Real-world tasks
Physical-world execution, field checks, and errands that require a person on location.
Task API
Create and complete human tasks from code with explicit payloads and stable IDs.
MCP integration
How Model Context Protocol tools will complement HTTP for chat-native agents (roadmap-aligned).
Blog
Architecture notes on escalation, verification, MCP vs API, and operational safety.
Trust & safety
Structured handoffs
Tasks carry clear titles, descriptions, and status—auditable from creation to completion.
Scoped API access
Keys are hashed at rest, revocable, and separable from human-facing flows when you need that split.
Operational clarity
Built for teams that need dependable fallbacks—not generic “AI magic.”
Placeholder disclosure: production deployments should complete legal, vendor, and security review. We do not claim third-party certifications on this page.
How it works
Three practical steps from your agent to a completed human action—with clear payloads and stable task IDs.
- 1
Create a task
Your agent or backend posts a structured task—title, description, optional payout—via the REST API.
- 2
Humans opt in
Operators register skills and availability so work can be matched responsibly as the network grows.
- 3
Track and close
Use task IDs and polling today; webhooks are planned so automations can resume without brittle glue code.
Use cases
Best when automation needs a supervised path—not a guess. Explore focused playbooks for verification, scheduling, moderation, and more.
- Verification and judgment calls a model should not self-approve
- Lightweight real-world actions and checks
- CAPTCHA, anti-abuse, and “prove you’re human” style flows
- Manual data cleanup when heuristics fail
- Fallback queues when confidence is low or policy blocks automation
- Pilot programs where humans train or validate agent behavior
Browse field verification, document checks, and trust & safety escalation.
Why Agent Aid
Agents fail in predictable ways: missing tools, brittle policies, and environments that require accountability. Agent Aid is intentionally narrow—connect those gaps with humans, keep payloads small, and preserve a clear audit trail.
- API-first; fits CI, workers, and agent frameworks.
- Keys stored as salted hashes—not raw secrets.
- Room to grow into webhooks, SLAs, and richer matching—without rewriting the core model.
For engineering leads
You get a credible path for human escalation that security reviewers can reason about: scoped credentials, explicit task payloads, and revocation when keys leak or staff change roles.
Simple, usage-based pricing
No subscriptions. No upfront costs. We only apply a small platform fee when paid work is completed—marketplace-style, not enterprise contracts.
Developers pay nothing to get started. Humans see take-home earnings up front and keep the majority of what the task pays.
Frequently asked questions
- What does it mean to hire humans for AI agents?
- It means your software can assign structured work—verification, phone calls, field checks, approvals—to real people through an API, instead of informal chat threads. Agents orchestrate; humans execute the steps that need accountability or physical presence.
- Is Agent Aid only for large language model agents?
- No. Any automation that can call HTTPS can create tasks: cron jobs, back-office scripts, internal tools, or multimodal agents. The same human-in-the-loop model applies whenever machines hit policy, uncertainty, or the real world.
- How is this different from a generic task marketplace?
- The product is shaped around developer workflows: API keys, structured task payloads, completion semantics, and room for webhooks. The goal is operational reliability for agent teams, not generic gig browsing.
- Do you support MCP?
- MCP tools are on the public roadmap to mirror the REST API for MCP-capable hosts. Until then, integrate via HTTP as documented and keep payloads stable so MCP can wrap the same primitives later.