InferWeave is an enterprise-grade platform for designing, deploying, and managing multi-agent workflows. Coordinate teams of AI agents to automate complex business processes with built-in human oversight and full audit trails.
MIT Licensed · Changelog · Launching Q3 2026
from inferweave import Agent, Workflow
# Define specialized agents with roles and tools
researcher = Agent(
name="researcher",
role="Gather and summarize market data",
tools=["web_search", "doc_reader"],
memory=True
)
analyst = Agent(
name="analyst",
role="Evaluate findings and produce report",
tools=["calculator", "chart_gen"],
memory=True
)
# Build the workflow with handoffs
workflow = Workflow(
name="market-analysis",
steps=[researcher, analyst],
handoff="sequential",
human_approval=True # gate before final output
)
result = workflow.run(input="Q2 competitor landscape")A complete platform for multi-agent orchestration. Define agent roles, design task dependencies, enforce human oversight, and monitor everything in production.
Design multi-step agent workflows in a drag-and-drop canvas or in code. Define task dependencies, conditional branches, parallel execution lanes, and convergence points. Export to workflow.yaml for version control.
Create specialized agents with distinct roles, tool permissions, and memory scopes. Assign agents to tasks based on capability matching. Manage agent configurations centrally with agents.yaml.
Insert approval gates at any point in a workflow. Reviewers receive structured summaries and can approve, reject, or redirect tasks. Configurable escalation policies and SLA timers keep processes moving.
Every agent action, tool invocation, handoff, and human decision is logged with full provenance. Integrate with your identity provider via SAML/OIDC for SSO. Role-based access control across all workflows and agents.
Go from idea to deployed multi-agent workflow in minutes. Configuration is declarative and version-controlled.
Create agents in agents.yaml with roles, tool access, memory settings, and guardrails. Each agent is a composable building block for your workflows.
Use the visual designer or write workflow.yaml to define task sequences, handoff protocols, parallel branches, and human approval gates.
Run inferweave deploy to push your workflow. Monitor agent activity, task completion rates, and bottlenecks from the real-time dashboard.
# workflow.yaml
name: customer-onboarding
version: 1
agents:
intake:
role: Extract and validate customer data
tools: [ocr, crm_lookup, validator]
compliance:
role: Run KYC/AML checks against regulations
tools: [kyc_api, sanctions_db]
provisioner:
role: Set up accounts and send welcome kit
tools: [account_api, email_sender]
steps:
- agent: intake
output: customer_profile
- agent: compliance
input: customer_profile
human_approval: true
sla_minutes: 30
- agent: provisioner
input: approved_profile
on_failure: escalate_to_humanTwo common patterns: defining agents with tools and memory in Python, and configuring human approval gates for sensitive workflows.
from inferweave import Agent, Tool, Memory
# Custom tool for the agent to use
crm_tool = Tool(
name="crm_lookup",
description="Search CRM by customer ID or name",
endpoint="https://internal.corp/api/crm",
auth="${CRM_TOKEN}"
)
support_agent = Agent(
name="support-tier-1",
role="Handle incoming support tickets, "
"classify urgency, gather context",
tools=[crm_tool, "knowledge_base"],
memory=Memory(
backend="redis",
ttl_hours=48,
scope="per_customer"
),
guardrails=["no_pii_in_logs", "max_actions_10"]
)
# Register and deploy
support_agent.register()# Multi-step procurement approval workflow
name: procurement-approval
version: 2
trigger: purchase_request_submitted
steps:
- name: extract_details
agent: doc_parser
action: Parse PO and extract line items
- name: vendor_check
agent: compliance_checker
action: Verify vendor against approved list
depends_on: extract_details
- name: manager_review
type: human_gate
assignee: ${requestor.manager}
sla_minutes: 120
escalate_to: vp_operations
depends_on: vendor_check
show_fields:
- total_amount
- vendor_name
- compliance_status
- name: execute_order
agent: procurement_agent
action: Submit PO to vendor system
depends_on: manager_reviewYou pay for InferWeave platform usage. Bring your own model provider keys — no markup on underlying LLM costs.