Bottega Engineering Orchestrator
Coordinate planning, implementation, review, and PR feedback across coding agents.
- Source repo
- vdaubry/bottega
- Stars
- ★ 91
- Last updated
- 1mo ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 62/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- An engineering team wants to coordinate planning, coding, and review for multiple users on a shared server.
- Not a fit if
- Developers who need a single-user local workflow
- Teams that cannot depend on external agent runtimes
- Maintenance
- Commits in the last 6 monthslast commit 2026-10-05
- Source review
- 62/100 · Some gaps
What does this agent do, and when should you use it?
Bottega is a multi-agent orchestration tool for engineering teams, delivered as a specification and a working reference implementation. The reference app uses a React, Vite, and Tailwind frontend, a Node.js, Express, and WebSocket backend, and SQLite for metadata and messages, with a multi-user web UI. It organizes work into planning, implementation, and review loops; the review agent runs QA scenarios and verifies the goal before a PR is opened. Teams can drive Claude Code, Codex, and OpenCode and assign different providers to different roles. Bottega is designed for remote server deployment, though it can also run locally; GitHub webhooks can trigger agents to address PR comments and keep CI green.
After a user submits a development task, a planning agent fills a plan template for human review and approval, then an implementation agent changes the project code. An adversarial code-review agent checks the code, runs the test scenarios that define the goal, and verifies completion; if the goal is unmet or errors occur, work returns to implementation for another loop. Once complete, the system opens a PR and tracks CI. The human reviews it on GitHub, where comments can reach a PR-feedback agent through a GitHub webhook so it can address them and keep CI green. Teams can also use Explore to have an agent read a plan and cited code and produce an interactive HTML page, or use Epics to turn a functional specification into a sequence of tasks.
- An engineering team wants to coordinate planning, coding, and review for multiple users on a shared server.
- A team wants to assign different coding agents to planning, implementation, review, and PR management on one task.
- A team maintaining a large codebase wants agents to iterate against QA scenarios that define the goal.
- A team wants GitHub PR comments to trigger an agent that addresses feedback and keeps CI green.
- A team needs to break a large feature specification into ordered tasks that converge on one PR.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
npm install -g pnpm@11
- Ports
5173·3001- Environment variables
JWT_SECRET- Config files
.env.example.env- Runtime
- Node.js 18+
How do you install or deploy this agent?
The reference implementation is in reference/. It requires Node.js 18+, pnpm 11, and at least one Claude Code, Codex, or OpenCode runtime; the provider CLI must be installed on the host. Clone the repository, enter the reference app, install dependencies, and copy the environment template:
git clone https://github.com/vdaubry/bottega.git
cd bottega/reference
npm install -g pnpm@11
pnpm install
cp .env.example .envSet a random JWT_SECRET in .env, then run onboarding and start the development server:
openssl rand -hex 64
pnpm onboarding
pnpm devOnboarding creates an admin account and seeds a sample project. The frontend is at http://localhost:5173 and the backend at http://localhost:3001.
How do you use this agent?
Before the first chat or agent run, connect at least one provider in the UI. Claude Code and Codex use an OAuth URL generated by the app; authorize in a browser and paste the returned code. OpenCode requires a Zen API key pasted into its panel. Each teammate connects their own provider account because credentials are stored per user. Then add or select a Git repository in the web UI, submit a task, and review and approve the generated plan; the system runs the implementation and review loop before opening a PR. Comments left on GitHub can trigger follow-up work through the webhook.
What are this agent's strengths and limitations?
- The specification is the primary product, so teams can implement their own version and select optional features.
- It supports Claude Code, Codex, and OpenCode, allowing different providers to serve different task roles.
- Its remote-first, multi-user design lets a server respond to GitHub PR comments through webhooks.
- The reference app requires Node.js 18+, pnpm 11, and at least one provider CLI installed on the host.
- Initial setup includes an environment file, JWT_SECRET, admin onboarding, and provider authentication.
- The reference implementation uses TypeScript, React, Express, and SQLite; teams using another stack must implement the specification themselves.
How does this agent compare with similar options?
The README names Conductor, GasTown, gstack, SpecKit, and Symphony as tools with overlapping workflows. Bottega highlights its remote-first multi-user deployment, support for Claude Code, Codex, and OpenCode, and a minimalist interface built around the typical web development workflow.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Bottega Engineering Orchestrator This agent | 62 · Some gaps | Web app | ★ 91 | 1mo ago | TypeScript | Codex · Claude Code |
| Agent Skills — Production-Grade Engineering Skills for AI Agents | 67 · Some gaps | CLIFree + model costs | ★ 119k | 2d ago | JavaScript | Codex · Claude Code · OpenAI API |
| Cezar — AI Coding Agent Orchestrator | 62 · Some gaps | CLIFree | ★ 1.1k | today | TypeScript | Codex · Claude Code |
| Plannotator | 80 · Good | CLIFreemium | ★ 14k | today | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The README describes human approval of plans and human review of the final PR, providing meaningful confirmation gates; the overview does not establish that every consequential operation requires confirmation, so user confirmation scores 2. It says provider credentials are stored per user, and the security policy explicitly covers authentication and credentials; the sampled code does not show storage or access-control details, so sensitive-data handling scores 2. Webhooks, OAuth, Zen API keys, server-side subprocesses, and repository operations are described, giving partial data-flow visibility. The effects are broad, while the sample does not explain their permission limits or recovery; least privilege, external effects, and rollback each score 1. CI pins some GitHub Action commits, but dependency inventory and vulnerability-response evidence are limited, so dependency security scores 2. Repository ownership and the MIT license are attributable, but publisher identity is unverified and maintenance responsibility is only partly defined by the security and contribution policies, so source attribution scores 1.
The README distinguishes the specification from the reference implementation and says the specification takes precedence, reducing ambiguity about their roles. The test workflow, boundary test, and CLI error handling offer some corroboration, but this sample cannot establish overall behavior, so self-consistency scores 2. Setup instructions identify Node, pnpm, and at least one agent runtime. CI uses a lockfile and pinned action commits, but third-party CLIs and runtimes remain externally supplied, so dependency availability scores 2. The CLI reports unknown commands and caught startup errors, and the server entry handles a missing .env file; coverage shown is limited, so failure messages score 2.
The materials identify engineering teams, remote shared deployment, nontechnical collaborators, and planning, implementation, review, PR feedback, and epic workflows, so audience and scenarios score 3. Multiple runtimes, optional extra specifications, and a replaceable technology stack show some extensibility, but the sample lacks a full capability matrix or explicit unsupported-scope list, so capability boundaries score 2. The README gives human approval points, provider setup, and GitHub webhook scenarios, but not a complete trigger-condition inventory, so trigger precision scores 2. Setup covers local startup and the remote-first design, while production deployment, network exposure, and operations guidance are limited in the sample, so environment fit scores 2.
The README has a clear specification index, core/optional split, reference-implementation explanation, and contribution guidance, so information architecture scores 3. Prerequisites, installation, environment variables, onboarding, provider authentication, and startup are specific, so install notes score 3. The sample uses Bottega and cloudcli names and several product layers without fully explaining naming relationships, so naming stability scores 1. The README includes a demo, getting-started example, and practice write-up, but no FAQ section, so examples and FAQ score 2. The security policy defines scope, supports only latest main, and states that security fixes are not backported; these are useful limits, though product operating limits are incomplete, so known limitations score 2. The LICENSE grants MIT terms, scoring 3. The supplied version information and support policy do not provide a clear release history or changelog, so versioning and changelog score 1. Contribution policy prioritizes specs and accepts reference bug fixes, while the security policy says only latest main is supported; publisher identity and maintainer path remain unclear, so maintenance responsibility scores 2.
The product organizes planning, agent execution, review, and PR feedback into a readable workflow with a clear intended output. The sample does not show concrete output formats or quality thresholds, so output usability scores 2. Multi-agent, multi-runtime coordination and interactive plan review offer identifiable workflow value, but the benefits are largely project claims without comparative evidence, so marginal value scores 2. A remote multi-user server and multiple agent processes carry deployment, runtime, and maintenance costs; the materials provide no cost estimate or scale guidance, so cost-benefit scores 1.
README claims about setup, workflow, and design have partial counterparts in the code, test configuration, and security policy, but many features are outside the supplied sample, so claim traceability scores 2. The README, security policy, CI, and test samples corroborate some aspects of server operation, testing, and maintenance, but do not independently corroborate all product claims, so cross-source corroboration scores 2. The README distinguishes the specification from the reference implementation, but claims about production use and benefits remain publisher assertions, so fact-inference separation scores 2.
- The product starts coding-agent subprocesses on a server and operates on repositories. Review deployment permissions, repository access, and webhook configuration before connecting it; this static sample does not establish those control details.
- The security policy supports only latest main and does not backport fixes. Running third-party agent CLIs also depends on their installation and authentication state.
FAQ
Does Bottega cost money to use?
Can I run it only on my own computer?
Does each teammate need to connect a provider?
Is this a single finished product to adopt as-is?
Is Bottega Engineering Orchestrator open source?
Is Bottega Engineering Orchestrator safe to use?
Is Bottega Engineering Orchestrator still maintained?
Who makes Bottega Engineering Orchestrator?
Thanks — we use this to improve the page.