AiMatch

A2Flow

A2Flow turns team procedures into repeatable workflows that an AI agent plans and executes with human approval gates. IT and operations teams can run those workflows in a shared chat with live task progress.

Visit A2Flow Compare alternatives Own this tool? Claim it
A2Flow screenshot

Who it's for

  • IT service teams that want to run repeatable service-request, change-execution, or announcement procedures.
  • Operations teams that need an approver to authorize consequential tool calls during a workflow.
  • Teams that maintain procedures as SKILL.md files in Git and want an agent to execute the published task graph.

How it fits your workflow

An author describes the intended procedure in a design chat. The agent reads the relevant Agent Skill, drafts a task graph with dependencies, and registers task templates. The team can adjust the plan before publishing it.

A user starts a published workflow in its own session. The agent asks for required information through forms in the chat and advances tasks in dependency order using only the MCP tools bound to each task.

When a tool call would change something, the run pauses for the designated approver. The approver decides in the same conversation; the agent then continues and reports task progress and results.

Pricing

The vendor doesn't publish prices. Check with A2Flow directly.

Prices checked on Nov 3, 2026 from the vendor's site. They can change; confirm before you buy.

Key features

Agent Skills stored in Git
Workflow procedures are SKILL.md files in Git repositories. The page describes cloning them on first run and treating them as versioned, reviewable, readable team procedures.
Reusable task-graph workflows
In a design session, the agent turns an intent into task templates with explicit dependencies and registers them as a DAG. Cycles are rejected, and a published workflow can be run repeatedly.
Human approval gates
Publishing a workflow approves its plan. During execution, calls that change something pause for a designated approver, who can approve, reject, or return the request in the chat.
Task-scoped MCP tools
MCP servers can be registered and tools bound to tasks during planning. At run time, each task can call only the tools bound to that task.
Forms and rich chat UI
The agent renders buttons, forms, and cards inline in the chat using A2UI; user interactions are returned to the workflow as tool results.
Live task progress and workflow views
Tasks progress through pending, in-progress, and completed states. Users can view the workflow as a sortable table or an automatically laid-out graph.

Works with

  • Agent Skills in Git repositories
  • MCP servers
  • Gemini, OpenAI, and Anthropic
  • PostgreSQL and SQLite

Limitations to know

  • The documented local setup requires Docker Compose and a Google API key for the default Gemini configuration.
  • Only the designated approver can resolve an approval request; the page says other users receive a 403.
  • At run time, a task can call only the MCP tools bound to that task.

Alternatives to consider

  • AilaFlow

    Choose it if you want an AI workspace where administrators configure access, available actions, and approval points for team workflows run through chat and forms.

Compare A2Flow side by side

Getting started

docker compose up --build

Docker Compose and a Google API key for the default Gemini configuration.

  1. Clone the repository and enter the project directory.
  2. Create an .env file containing GOOGLE_API_KEY=your_google_api_key_here.
  3. Run docker compose up --build to start the PostgreSQL, backend, and frontend stack.
  4. Open http://localhost:3000.