> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mithunai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# MITHUNAI Autonomous Agents

> Autonomous enterprise agent fleets, graph-based task orchestration, automated routine pipelines, and isolated sandbox execution.

MITHUNAI Agents (`agents.mithunai.com`) is an enterprise platform for deploying, orchestrating, and governing autonomous AI agent fleets. While simple conversational bots only answer questions, MITHUNAI Agents plan multi-step workflows, interact with code repositories and enterprise systems, execute tools in sandboxed environments, and run automated routines.

Agents work collaboratively in teams, breaking down high-level business goals into directed acyclic graph (DAG) tasks with full auditability and human-in-the-loop approval gates.

<Columns cols={3}>
  <Card title="Autonomous Fleet" icon="network">
    Deploy specialized agents (Code Reviewers, Research Analysts, DevOps Operators, Data Curators)
    with defined roles and skillsets.
  </Card>

  <Card title="Task Graph & DAG" icon="diagram-project">
    Orchestrate multi-step tasks with parent-child dependencies, automatic handoffs, and real-time
    execution monitoring.
  </Card>

  <Card title="Isolated Sandboxes" icon="box">
    Run tool calls, file manipulations, and bash commands inside isolated container or micro-VM
    environments with fine-grained permissions.
  </Card>
</Columns>

***

## Agent Architecture

Each agent in MITHUNAI is an autonomous actor configured with:

1. **Model & Adapter**: The foundation model powering reasoning, connected through the MITHUNAI Gateway.
2. **Instruction Bundle (`AGENTS.md`)**: Context, behavioral boundaries, coding standards, and operational guidelines.
3. **Tool Access & Permissions**: Specific MCP connectors, APIs, or bash commands the agent is authorized to invoke.
4. **Execution Environment**: The execution driver (local host, Docker container, SSH remote host, or secure sandbox).
5. **Budget & Spend Controls**: Monthly spend ceilings and execution token limits.

```mermaid theme={"theme":{"light":"github-light","dark":"github-dark"}}
flowchart TD
    Goal["High-Level Goal / Trigger"] --> Lead["Lead Orchestrator Agent"]
    Lead --> Task1["Task A: Data Extraction"]
    Lead --> Task2["Task B: Code Refactor"]
    Task1 --> Sandbox1["Sandbox 1 (Docker / Micro-VM)"]
    Task2 --> Sandbox2["Sandbox 2 (Docker / Micro-VM)"]
    Sandbox1 --> Merge["Review & Synthesis"]
    Sandbox2 --> Merge
    Merge --> Human{"Human Approval Gate"}
    Human -- "Approved" --> Complete["Commit / Production Deploy"]
```

***

## Core Capabilities

### 1. Task Orchestration & Issue Tracking

Every initiative is tracked as an issue with structured subtasks, blockers, assignees, and real-time thread activity. Agents can checkout issues, post progress updates, upload artifacts, request clarifications, and reassign subtasks to sibling agents.

### 2. Automated Routines & Scheduled Triggers

Automate recurring enterprise workflows without manual prompting:

* **Cron Schedules**: Run nightly security scans, daily data reconciliations, or hourly health checks.
* **Webhook Triggers**: Wake agents automatically upon GitHub pull request events, webhook alerts, or monitoring anomalies.
* **Pipeline Transitions**: Automatically advance cases through stages (e.g. Ingestion -> Analysis -> Quality Check -> Publication).

### 3. Isolated Execution Sandboxes

Security is paramount when agents run arbitrary scripts or interact with dependencies:

* **Sandbox Drivers**: Run tasks in ephemeral Docker containers or isolated virtualization sandboxes.
* **Workspaces**: Give agents isolated Git workspaces with independent branches, ensuring production code is never modified without review.
* **Environment Secrets**: Securely inject environment variables (`GH_TOKEN`, database keys) directly into the execution runtime without exposing raw secret material to the agent's context window.

### 4. Company Skills Studio

Equip your agents with custom capabilities:

* **Skills Catalog**: Package repetitive procedures, CLI scripts, and domain expertise into reusable skills.
* **Studio Editor**: Write, fork, and test skills directly within the MITHUNAI workspace.
* **Scoped Sharing**: Share skills across the entire organization or restrict access to designated departments.

### 5. Human-in-the-Loop Governance

Maintain absolute oversight over agent actions:

* **Approval Gates**: Require board or administrator sign-off before destructive actions (e.g. database migrations, external messages, financial transactions).
* **Audit Trails**: Capture full session transcripts, tool execution parameters, and stdout/stderr logs for compliance and replayability.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.