# Alfrig

> An office of AI agents that turns a request into a reviewed, tested and merged pull request. Works with GitHub, GitLab, Bitbucket and Azure DevOps, and the tools your team already uses. Start free.

Alfrig is an office of AI agents for software teams. Agents are hired like teammates (name, role, department), take a task from request to merged pull request, and every step is visible in a pixel-art office and in a detailed log. Web app at https://app.alfrig.com/ and desktop app for macOS, Windows and Linux. Site in English, Portuguese and Spanish.

## How it works

1. **Connect your repositories and a model key**: Import repositories straight from your account. Alfrig maps the catalog, the stack and the dependencies between projects on its own.
2. **Hire agents**: Give them a name, a look, a department and a role. With two agents, one builds and the other reviews.
3. **Describe the task**: In a chat with Alfrig, a form, a label in Jira, Linear or ClickUp, through the API or via MCP.

The pipeline: Selection (The agent picks the projects from the catalog when the task does not say.) → Investigation (Reads the repositories and writes the spec: what changes, where and how to verify.) → Development (Writes the code, runs the tests and opens one PR per repository.) → Review (Another agent reads the diffs, runs the tests and checks every criterion.) → Merge (With green CI, squashes in dependency order and deletes the branches.) Changes requested in review or a red CI go back for fixing on the same branches, up to 3 rounds.

## Why teams use it

- **Reviewed by another agent**: The author never approves its own PR. Every acceptance criterion in the spec is checked; if one fails, approval is withdrawn.
- **Spec before code**: Every task starts with a read-only investigation: problem, acceptance criteria, test plan and what is out of scope.
- **Asks instead of guessing**: When information is missing, the agent asks before coding, with quick answers by link, no account needed.
- **Many repositories, in parallel**: Backend and frontend in one task: one agent per repository, and the merge only happens when every CI is green.
- **A complete log**: Every step keeps the prompt, commands, files, tests, time and cost. You can see exactly what the agent did.
- **Any language**: Node, PHP, Python, Java/Kotlin, .NET, Go, Ruby and mixed projects. Monorepos included.

## Control

Every step runs on its own or waits for a person. Auto mode on, it goes from request to merge with nobody pushing; off, it stops wherever you want. Plan approval: The task pauses after the investigation and waits for someone to approve or adjust the plan. Auto review: Every PR opened by an agent goes straight to another agent’s review. Auto merge: Approved in review with green CI, the merge happens right away. Resolve merge conflicts: If the base moved, an agent merges it and resolves the conflict on its own.

## Security

- **Nothing lands without review**: Merge only after an approved review and green CI. Auto-merge and plan approval can require a person.
- **The AI never sees your credentials**: Clone, push, PR and merge are done by Alfrig. The agent works on a copy without the repository token.
- **Isolated, disposable environment**: Each step runs on its own machine with no inbound connections, destroyed at the end. Tests run offline.
- **A database of your own**: Each organization has its own database. Keys encrypted with AES-256-GCM using a key unique to your organization.
- **Role-based permissions**: Owner, admin, member and guest. Connected apps only get what the person’s role allows.
- **Deletion that means it**: The owner can delete the organization for good: database, files and encryption key. Technical logs kept for 12 months.

## Pricing

Unlimited members on every plan. Model usage is paid directly to the provider you choose.

- **Free** — US$ 0. Watch one agent build while another reviews. 2 agents; 20 tasks a month; 5 machine hours per agent; Your own model keys; Unlimited members.
- **Team** — US$ 15 per agent/month. Get the backlog moving. As many agents as you hire; Unlimited tasks; 20 machine hours per agent; Everything in Free; Email support.
- **Enterprise** — Custom. For multi-product groups and IT requirements. Unlimited agents and tasks; No machine-hour allowance; Dedicated infrastructure; SSO and access policies; Invoicing, contract, SLA and priority support.

The machine-hour allowance is a notice, not a block: overage shows up in the usage report.

Model costs (real examples): Add formatCpf utility and export: US$ 0.02; The README doesn’t explain how to run the project on a newcomer’s machine: US$ 0.17; Add a local setup section to the README: US$ 0.14; Map a 4-project catalog: US$ 0.16; Survey the frontend design system: US$ 0.31; The project’s AGENTS.md: US$ 0.05; Read the project’s code: US$ 0.04. Bug fixes in large systems, with deep investigation on the strongest model, came in between US$ 5 and US$ 12. Cost varies with the chosen model and the size of the codebase.

## Integrations

GitHub (and GitHub Enterprise), GitLab (repositories and merge requests), Bitbucket (repositories and pull requests), Azure DevOps (Repos and pull requests), Jira (labels, comments, subtasks), ClickUp (tags, comments, subtasks), Linear (labels, states, sub-issues), Public API (OpenAPI, scoped keys), Webhooks (signed events), MCP (25 tools, OAuth 2.1), Desktop app (Mac, Windows and Linux). A tag opens the task; progress comes back as comments and status. Questions and approvals are answered right in the comment. Coming soon: GitHub Issues.

## Beyond engineering

- **Agents with a name, a role and a face**: Hire, pick the department and design the look. Talk to the agent on a task like you would to a colleague.
- **Learning per project**: Lessons saved after every task and an AGENTS.md kept by the agents. The next task already knows.
- **Product department**: A broad request becomes an initiative: the PM analyzes and prioritizes, the PO documents, the designer designs, and tasks flow to Engineering.
- **Strategy with OKRs**: A yearly plan built in a chat, spread across quarters, with check-ins. The PM prioritizes by it.
- **Shareable artifacts**: Specs and prototypes with versions, in a searchable gallery. Private, organization or public: you choose.
- **Mapping, design system and desktop**: Alfrig reads your projects, surveys your design system and opens in its own window on Mac, Windows and Linux.

## FAQ

### What is Alfrig?

An office of AI agents that turns a request into a reviewed, tested and merged pull request. One agent investigates the code, writes the spec, builds and tests; another agent reviews it against every acceptance criterion; the merge only happens with green CI. You follow everything in a pixel-art office and in a record of every step.

### How much does Alfrig cost?

The Free plan costs nothing: 2 agents and 20 tasks a month, no card. The Team plan is US$ 15 per agent per month, with unlimited tasks. The Enterprise plan is priced on request. Every plan has unlimited members. Model usage is paid directly to the provider (Anthropic, OpenAI or z.ai), with no markup from Alfrig.

### Is the code the agent writes any good?

Every task starts with a spec and acceptance criteria. Tests related to the change run before the PR, and another agent reviews against each criterion, with up to 3 fix rounds. If tests fail, the work becomes a draft PR, never a merge.

### What access to my repositories does Alfrig need?

A token with access to the repositories you choose. On GitHub, a fine-grained token limits it to the ones you authorize. The token is encrypted and never handed to the agent.

### Which models can I use?

Anthropic (Claude), OpenAI (GPT) and z.ai (GLM), with your organization’s key. New models your key can reach show up for selection automatically.

### What if the agent gets it wrong?

Request a fix from the current code, even by re-adding the tag after QA rejects it. Stalled tasks resume where they stopped, with new guidance.

### Is my code used to train models?

Alfrig does not train models. Data use depends on the provider you connect; check its policy.

### Is auto-merge mandatory?

No. It is on by default, but only happens after an approved review and green CI. You can turn it off in the workflow, along with requiring a person to approve the plan.

### How much does a task cost in model usage?

It depends on the model and the size of the code. Real examples from September 2026: US$ 0.02 to add a function with tests, US$ 0.05 to write a project’s AGENTS.md, US$ 0.31 to survey a frontend’s design system. Bug fixes in large systems, on the strongest model, came in between US$ 5 and US$ 12. Every task shows what it cost.

### Does it work with GitLab, Bitbucket or Azure DevOps?

Yes. Alfrig works with GitHub (including GitHub Enterprise), GitLab, Bitbucket and Azure DevOps. Tasks can come from Jira, Linear, ClickUp, a chat, the API or MCP.

### Which languages does it support?

Node, PHP, Python, Java/Kotlin, .NET, Go, Ruby and projects that mix more than one, monorepos included. A single task can change several repositories at once, with one agent per repository.

### How do I open a task?

In a chat with Alfrig, in a form, with a label in Jira, Linear or ClickUp, through the public API or through MCP, from inside your AI assistant. Progress comes back as comments and status in the tool it came from.

### How is it different from a coding assistant in the editor?

An editor assistant helps whoever is coding, line by line. Alfrig takes the whole task and runs the cycle: it investigates, writes the spec, builds, tests, asks another agent to review and merges with green CI, with nobody sitting in the editor. You decide which steps need a person’s approval.

### Does Alfrig do more than write code?

Yes. Beyond Engineering there is a Product department, with a PM, a PO and a designer as agents: a broad request becomes an initiative that is analyzed, prioritized, documented and designed before it reaches Engineering. And there is Strategy: Alfrig builds OKRs and KPIs through a chat, and the PM prioritizes by them.

### How do agents learn from each task?

When a task is done, the agent saves short lessons about the project that feed the context of the next tasks; you review, write and delete whatever you want. The project’s AGENTS.md is written when missing and updated by PR. Skills per department and step apply your company’s standards.

### Does the designer use my product’s design system?

Yes. The designer surveys your frontend’s tokens, libraries, patterns and components from the code and checks each sample in a browser. The screens on the design board come out with that system and your brand rules, checked in a browser.

### Who can I share specs and prototypes with?

Each artifact can be private (only whoever chatted with the agent), organization-wide (anyone with a login) or public (anyone with the link, no account). Artifacts have versions and live in a searchable gallery.

### How does Alfrig know which repository to change?

When you import repositories, an agent reads each project, writes its card (what it does, stack and how to test it) and works out the dependencies between them, each with the snippet that proves it. That map is how agents pick where to change code and in which order to merge.

### Is there a desktop app?

Yes, for Mac (Apple Silicon and Intel), Windows and Linux (AppImage). It opens Alfrig in its own window and updates itself in the background.

## Links

- [Home (English)](https://alfrig.com/en/)
- [Features (English)](https://alfrig.com/en/features/)
- [Interactive demo (English)](https://alfrig.com/en/demo/): click through a real task from request to merge
- [Home (Português)](https://alfrig.com/)
- [Funcionalidades (Português)](https://alfrig.com/funcionalidades/)
- [Inicio (Español)](https://alfrig.com/es/)
- [Funcionalidades (Español)](https://alfrig.com/es/funcionalidades/)
- [Start free](https://app.alfrig.com/)
- Contact: contato@alfrig.com