Fulling
AI-powered full-stack engineering agent with per-user Kubernetes credential boundaries for persistent workspaces.
- Source repo
- FullAgent/fulling
- Stars
- ★ 2.4k
- Last updated
- 2mo ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 48/100 · Major gaps
At a glance
- How it runs
- Works with
- Platform-specific
- Cost
- Free software; you pay for model usage
- Setup effort
- High · needs real infrastructure
- You'll need
- Typical use
- Developers wanting an AI coding agent with isolated Kubernetes credentials per user.
- Not a fit if
- Users upgrading from v2; this release does not migrate v2 data
- Teams unwilling to store Kubernetes credentials as plaintext in the database
- Source review
- 48/100 · Major gaps
What does this agent do, and when should you use it?
Fulling is an AI-powered full-stack engineer agent built on Next.js, Claude, shadcn/ui, and PostgreSQL. The v3 foundation focuses on identity and Kubernetes credential boundary, providing GitHub-only sign-in via Better Auth, PostgreSQL-backed user management, and a protected workspace entry. Each user stores a plaintext kubeconfig that is validated through Kubernetes SelfSubjectReview, enabling a user-scoped Kubernetes client boundary. The public application can run without PostgreSQL or OAuth, while the legacy authenticated workspace requires them. This version uses a new baseline schema and does not migrate v2 data.
Fulling performs GitHub OAuth authentication using Better Auth, manages users and sessions through Prisma and PostgreSQL, stores and validates per-user kubeconfigs via Kubernetes SelfSubjectReview, and provides a user-scoped Kubernetes client boundary. It rejects insecure kubeconfig configurations including executable plugins, local file credential fields, proxies, and non-HTTPS API servers. The public app starts with sign-in disabled, while the authenticated workspace requires a database and GitHub OAuth app.
- Developers wanting an AI coding agent with isolated Kubernetes credentials per user.
- Teams needing to manage multiple users' access to Kubernetes clusters securely.
- Developers building a Next.js and PostgreSQL-based authenticated workspace quickly.
- DevOps engineers verifying kubeconfig security before deployment.
- Operators needing to clean up Kubernetes resources from v2 before upgrading.
How do you install or deploy this agent?
Clone the repo, install Node.js 24, and run npm ci. The public app runs without environment variables; use npm run dev. For the authenticated workspace, copy .env.template to .env.local, fill database, Better Auth, and GitHub OAuth values, then run npm run prisma:migrate and npm run dev.
How do you use this agent?
Start the development server for the public app. For sign-in, configure the GitHub OAuth callback as ${BETTER_AUTH_URL}/api/auth/callback/github. For deployment, follow the Vercel guide for zero-configuration builds.
What are this agent's strengths and limitations?
- Provides a dedicated Kubernetes credential boundary per user.
- Robust kubeconfig validation using Kubernetes SelfSubjectReview, rejecting risky configurations.
- Modern stack (Next.js, Prisma, shadcn/ui) facilitates extension.
- Clear architecture and deployment documentation.
- GitHub-only sign-in limits identity provider options.
- Kubeconfigs stored in plaintext pose a risk if the database is compromised.
- No migration path from v2; requires a fresh database.
- Full workspace functionality depends on Kubernetes and PostgreSQL.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Fulling This agent | 48 · Major gaps | Web appFree + model costs | ★ 2.4k | 2mo ago | TypeScript | — |
| CloudBase AI Toolkit | 59 · Major gaps | Agent plugin / skillFreemium | ★ 1.1k | today | TypeScript | Codex · Claude Code |
| Open Generative UI | 44 · Major gaps | CLIFree + model costs | ★ 3.5k | 5d ago | TypeScript | Claude API |
| Open Multi-Agent Canvas | 43 · Major gaps | Web appFree + model costs | ★ 551 | 4mo ago | TypeScript | — |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: README explicitly states kubeconfigs are stored as plaintext in PostgreSQL and lists validation rejections for plugins, proxies, etc., indicating basic sensitive data handling. However, there is no user confirmation mechanism (e.g., confirmation before sensitive operations), external effects (like SSRF boundary) are mentioned but implementation details are not provided. Dependency security has npm audit scripts but no audit results. Rollback only mentions deployment reset docs, no concrete rollback mechanism. Source attribution relies only on README and package.json author fields, not verified. Therefore deductions: user_confirmation, external_effects, rollback, source_attribution lack evidence.
Evidence shows: README and package.json scripts and dependencies are consistent, CI config includes lint, test, build, indicating some self-consistency. Dependencies are pinned or use ^, but no lockfile or integrity checks. For failure messages, README does not provide common error handling guidance. Therefore deduction: failure_messages lacks evidence.
Evidence shows: README clearly targets AI full-stack engineers and describes use cases (development, deployment). Capability boundaries are described in Credential Boundary section, but trigger precision (e.g., command trigger conditions) is not detailed. Environment fit supports Node.js 24 and Vercel deployment, but other environments are not mentioned. Therefore deduction: trigger_precision lacks evidence.
Evidence shows: README provides architecture doc link, installation steps, command list, clear information architecture. Install notes include local development and Vercel deployment. Naming stability: version is 3.0.0, but no changelog. Examples and FAQ missing. Known limitations mentioned in README (e.g., SSRF boundary). License is MIT. Maintenance responsibility not clear. Therefore deductions: naming_stability, examples_and_faq, maintenance_responsibility lack evidence.
Evidence shows: README describes product value, but no actual output examples or user feedback. Marginal value not quantified. Cost-benefit not discussed. Therefore deductions: output_usability, marginal_value, cost_benefit lack evidence.
Evidence shows: Claims in README (e.g., security boundaries) are not backed by tests or verification results. Cross-source corroboration insufficient, relying only on single repository. Facts and inferences not clearly separated. Therefore deductions: claim_traceability, cross_source_corroboration, fact_inference_separation lack evidence.
- Kubeconfigs are stored as plaintext; database breach could expose Kubernetes credentials.
- README explicitly allows users to configure any HTTPS API server, posing SSRF risk.
- No user confirmation mechanism; sensitive operations may execute automatically.
- Dependency audit scripts exist but no audit results provided; cannot confirm no known vulnerabilities.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v2.0.0 Fulling v2.0.0
Release notes excerpt
Fulling v2.0.0 Fulling v2.0.0 is the release snapshot for the v2 control-plane work. This release is not intended to claim the full future platform surface. It stabilizes the current agentic sandbox foundation so main can move on to 3.0 development with a clean historical branch. Release Positioning Fulling v2 is a browser-based AI development environment with: - project creation from scratch - GitHub App based repository import - isolated Kubernetes sandboxes - browser terminal and file manager access - live HTTPS runtime URLs - optional PostgreSQL databases …View on GitHub ↗ -
v1.0.0 v1.0.0 - MVP Release
Release notes excerpt
Fulling v1.0.0 - MVP Release 🚀 Fulling v1.0.0 is here. This release wraps up our MVP phase and proves the core concept: an AI-driven full-stack engineer agent that actually works. We've completely overhauled the UI to match the IDEs you're used to, and validated a new runtime environment built on Kubernetes StatefulSets. Highlights Runtime & Terminal - Claude Code CLI: Native support within the runtime container (#3, #18). - Persistent Sessions: We switched to Kubernetes StatefulSets, so your terminal sessions and workspaces stick around even if you refresh (#27, #103). …View on GitHub ↗