DevelopersAutonomous Agents

Agent Framework

Autonomous RA Agents

Seven specialized agents that handle the full product compliance lifecycle. Each maps to a real-world regulatory affairs role, uses scoped MCP tools, and escalates to humans when confidence drops.

7Specialized Agents

12MCP Tools

HITLEscalation Built-in

Get API Key Full API Docs

Overview

Inlet's agent framework provides seven autonomous regulatory affairs (RA) agents that mirror the roles on a real compliance team. Agents run via the Claude API with scoped MCP tool access, enabling them to read and write platform data while respecting permission boundaries.

Pipeline Orchestration

Agents execute in a defined sequence, with each stage feeding the next.

Scoped Tool Access

Each agent only sees the MCP tools relevant to its role -- no over-permissioning.

Human-in-the-Loop

Escalation triggers pause the pipeline and surface issues for human review.

Agent Delegation

Agents can invoke other agents when the task crosses role boundaries.

Multi-Tenant Isolation

All agent actions are scoped to the requesting company -- no data leakage.

Full Audit Trail

Every action, delegation, and escalation is logged with timestamps and rationale.

How It Works

1You call POST /api/v1/agents/pipeline/run with a SKU ID and target markets.
2The Formulation Analyst agent picks up the SKU, validates ingredients, and passes it to the Compliance Analyst.
3Each agent runs its checks using its scoped MCP tools, creates tasks for any issues, and delegates to the next agent.
4If an escalation trigger fires, the pipeline pauses and notifies your team for human review (HITL).
5After QA approval, the pipeline completes and the SKU is ready for market submission.

Pipeline Flow

The default pipeline runs agents in sequence. Market Intelligence and Regulatory Strategist agents run independently or are invoked on-demand by other agents.

Default Pipeline

Formulation

Compliance

Label Review

Submission

Human Review

Market Intelligence

Independent / On-demand

Monitors regulatory databases continuously. Can be scheduled on a cron or invoked by the Regulatory Strategist when assessing a new market.

Regulatory Strategist

Independent / On-demand

Advises on market entry sequencing and regulatory pathways. Invoked ad-hoc or by the Compliance Analyst when multi-market strategy is needed.

Agent Profiles

Click any agent to expand its full profile, including allowed MCP tools, pipeline stages, task types, escalation triggers, and delegation targets.

API Reference

All agent endpoints require a valid API key with the Growth or Enterprise plan. Requests are scoped to your company.

GET/api/v1/agents/profiles

Returns all agent profiles with their capabilities, allowed tools, pipeline stages, and escalation triggers.

bash

curl https://api.inlet.run/api/v1/agents/profiles \
  -H "Authorization: Bearer ink_live_your_key_here"

POST/api/v1/agents/pipeline/run

Runs the complete autonomous RA pipeline for a SKU across one or more target markets. Returns a run ID you can use to poll for status.

curl -X POST https://api.inlet.run/api/v1/agents/pipeline/run \
  -H "Authorization: Bearer ink_live_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "sku_id": "550e8400-e29b-41d4-a716-446655440000",
    "target_markets": ["US", "EU"],
    "skip_stages": []
  }'

Response

json

{
  "run_id": "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
  "status": "running",
  "actions_taken": 0,
  "escalations": 0,
  "current_agent": "formulation_analyst",
  "started_at": "2026-04-13T10:30:00Z"
}

POST/api/v1/agents/run

Run a single agent ad-hoc outside the pipeline. Useful for targeted checks, re-running a specific stage, or building custom workflows.

curl -X POST https://api.inlet.run/api/v1/agents/run \
  -H "Authorization: Bearer ink_live_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "agent_role": "compliance_analyst",
    "sku_id": "550e8400-e29b-41d4-a716-446655440000",
    "instruction": "Check this SKU against EU cosmetics regulation",
    "target_markets": ["EU"]
  }'

GET/api/v1/agents/pipeline/runs/{run_id}

Poll for the status of a pipeline run. Returns the current agent, actions taken, escalations, and completion status.

bash

curl https://api.inlet.run/api/v1/agents/pipeline/runs/a1b2c3d4-e5f6-7890-abcd-ef1234567890 \
  -H "Authorization: Bearer ink_live_your_key_here"

Response (completed)

json

{
  "run_id": "a1b2c3d4-e5f6-7890-abcd-ef1234567890",
  "status": "completed",
  "actions_taken": 12,
  "escalations": 0,
  "current_agent": null,
  "started_at": "2026-04-13T10:30:00Z",
  "completed_at": "2026-04-13T10:32:15Z",
  "stages_completed": [
    "formulation_analyst",
    "compliance_analyst",
    "label_reviewer",
    "submission_coordinator",
    "quality_assurance"
  ],
  "results": {
    "compliance_status": "compliant",
    "markets_checked": ["US", "EU"],
    "tasks_created": 3,
    "labels_reviewed": 1
  }
}

Run Status Values

runningPipeline is actively executing agents

completedAll stages finished successfully

awaiting_humanPaused for HITL review (escalation triggered)

failedAn unrecoverable error occurred

escalatedEscalation sent to team, awaiting resolution

Configuration

Configure API access, HITL checkpoints, and custom instructions for agent runs.

API Key Setup

1Navigate to Settings → API Keys in your Inlet dashboard.
2Create a new key with the agents:run scope. Keys are prefixed with ink_live_.
3Store the key securely. It is shown only once at creation time. Agent access requires the Growth or Enterprise plan.

HITL Checkpoints

By default, escalation triggers pause the pipeline and notify your team. You can configure which triggers require human review and which should auto-resolve.

json

{
  "hitl_config": {
    "pause_on": [
      "novel_ingredient",
      "regulatory_ambiguity",
      "approval_gate"
    ],
    "auto_resolve": [
      "confidence_below_threshold"
    ],
    "notify_channels": ["slack", "email"],
    "timeout_hours": 48,
    "fallback_action": "escalate_to_owner"
  }
}

Custom Instructions

Pass custom instructions to any agent to override default behavior. Instructions are appended to the agent's system prompt for that run only.

python

resp = requests.post(
    "https://api.inlet.run/api/v1/agents/run",
    headers={"Authorization": "Bearer ink_live_your_key_here"},
    json={
        "agent_role": "label_reviewer",
        "sku_id": "550e8400-...",
        "instruction": (
            "Focus on EU allergen disclosure requirements. "
            "Flag any missing INCI names for fragrance components. "
            "Apply strict interpretation of Regulation 1223/2009 Annex III."
        ),
        "target_markets": ["EU"],
    },
)

Integration Patterns

Common patterns for embedding agents into your existing workflows.

Trigger Pipeline on SKU Creation

Use a webhook listener to automatically run the full pipeline whenever a new SKU is created or moves to "active" status.

# When you receive a sku.activated webhook, trigger the pipeline:
curl -X POST https://api.inlet.run/api/v1/agents/pipeline/run \
  -H "Authorization: Bearer ink_live_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "sku_id": "550e8400-e29b-41d4-a716-446655440000",
    "target_markets": ["US", "EU"]
  }'

Schedule Market Intelligence Runs

Run the Market Intelligence agent on a schedule to monitor regulatory changes and generate proactive alerts.

python

# Using APScheduler or any cron-compatible scheduler
from apscheduler.schedulers.blocking import BlockingScheduler
import requests

API_KEY = "ink_live_your_key_here"

def run_market_intelligence():
    requests.post(
        "https://api.inlet.run/api/v1/agents/run",
        headers={"Authorization": f"Bearer {API_KEY}"},
        json={
            "agent_role": "market_intelligence",
            "instruction": "Scan all monitored markets for regulatory changes in the last 24 hours.",
            "target_markets": ["US", "EU", "UK", "CA", "MY", "SG"],
        },
    )

scheduler = BlockingScheduler()
scheduler.add_job(run_market_intelligence, "cron", hour=6, minute=0)
scheduler.start()

Build a Custom Agent with MCP Tools

Use the MCP tool endpoints directly to build your own agent logic. This gives you full control over tool orchestration while leveraging Inlet's compliance data.

python

import anthropic
import requests

INLET_API = "https://api.inlet.run/api/v1"
API_KEY = "ink_live_your_key_here"

# Fetch available MCP tools
tools_resp = requests.get(
    f"{INLET_API}/mcp/tools",
    headers={"Authorization": f"Bearer {API_KEY}"},
)
inlet_tools = tools_resp.json()["tools"]

# Use tools with Claude
client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=4096,
    system="You are a regulatory compliance assistant.",
    tools=inlet_tools,
    messages=[{
        "role": "user",
        "content": "List all SKUs that are non-compliant in the EU market.",
    }],
)

# Execute tool calls via Inlet MCP
for block in response.content:
    if block.type == "tool_use":
        result = requests.post(
            f"{INLET_API}/mcp/execute",
            headers={"Authorization": f"Bearer {API_KEY}"},
            json={"tool": block.name, "arguments": block.input},
        )
        print(result.json())

Connect Agents to Slack Notifications

Inlet's Slack integration automatically posts agent escalations and pipeline completions to your configured channel. Install via Settings → Integrations → Slack.

Navi12:31 PM

Pipeline completed for Hydrating Serum SPF30 across US, EU. 12 actions taken, 0 escalations. Status: compliant.

Escalation Triggers

When an agent encounters a situation outside its confidence bounds, it fires an escalation trigger. The pipeline pauses and your team is notified.

Trigger	Description	Agents
`novel_ingredient`	Ingredient not found in master database or has no regulatory data	formulation_analyst
`confidence_below_threshold`	Agent confidence score falls below the role-specific threshold	formulation_analystcompliance_analystquality_assurance
`regulatory_ambiguity`	Regulation text is ambiguous or multiple interpretations are possible	compliance_analystregulatory_strategistlabel_reviewer
`conflicting_regulations`	Two or more regulations contradict each other for the same ingredient or claim	compliance_analystmarket_intelligencequality_assurance
`high_risk_market`	Target market has strict or rapidly changing regulatory requirements	regulatory_strategistmarket_intelligence
`approval_gate`	Pipeline stage requires explicit human sign-off before proceeding	submission_coordinatorquality_assurance
`customer_request`	Customer has specifically requested manual review or custom handling	label_reviewersubmission_coordinator

Handling Escalations

When a pipeline enters the awaiting_human state, your team can review the escalation, make a decision, and resume the pipeline.

bash

# Resume a paused pipeline after human review
curl -X POST https://api.inlet.run/api/v1/agents/pipeline/runs/{run_id}/resume \
  -H "Authorization: Bearer ink_live_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "resolution": "approved",
    "notes": "Reviewed novel ingredient -- acceptable for EU market per SCCS opinion 2025/03."
  }'

Autonomous RA Agents

Overview

Pipeline Orchestration

Scoped Tool Access

Human-in-the-Loop

Agent Delegation

Multi-Tenant Isolation

Full Audit Trail

How It Works

Pipeline Flow

Default Pipeline

Market Intelligence

Regulatory Strategist

Agent Profiles

Formulation Analyst

Compliance Analyst

Regulatory Strategist

Label Reviewer

Submission Coordinator

Market Intelligence

Quality Assurance

API Reference

Response

Response (completed)

Run Status Values

Configuration

API Key Setup

HITL Checkpoints

Custom Instructions

Integration Patterns

Trigger Pipeline on SKU Creation

Schedule Market Intelligence Runs

Build a Custom Agent with MCP Tools

Connect Agents to Slack Notifications

Escalation Triggers

Handling Escalations