We're a Berlin-based development house building EU sovereign AI infrastructure. You're using AI tools to build things, solve problems, automate tasks — but your agent forgets everything when you start a new session. We fix that, with GDPR, DORA, NIS2, and EU Cyber Resilience Act compliance built in from day one.

Most AI agents repeat the same mistakes and forget what worked. Agentic Loop is different — it follows your process, learns from failures, and gets better over time.

Why people use it:

• Follows your workflow — write down how you want tasks done, and it sticks to that process
• Learns from mistakes — every time something goes wrong, it figures out why and avoids it next time
• Works across all your projects — one agent handles everything, keeps each project separate
• You can see what it's doing — full activity logs, so when it breaks you know exactly why
• Add your own tools — extend it without touching the core code

6 trait crates, 5 implementation crates — trait-first, WASM-safe, git-refs storage.

nix develop --command cargo build
agentic-loop run quick "Fix the failing tests"

Read the docs | Marketing overview | Full feature list

Multi-tenant Git hosting with built-in file storage, user permissions, and a plugin system — all in a single program. Built to support AI agent workflows.

What it gives you:

• Host Git repositories — push, pull, and clone over SSH or HTTP
• Large file support — handles datasets, models, and big files with quotas and S3 backup
• User management — control who can access what, with role-based permissions
• REST API — manage repos, users, and settings programmatically
• Enterprise features available — backup, compliance (DORA/NIS2/GDPR), replication, CI/CD, security scanning, webhooks (commercial licence)

nix build
./result/bin/agentic-repos init
./result/bin/agentic-repos

Read the docs | API reference | Deployment guide

If you're using Claude Code, OpenCode, pi, or OpenClaw, your agent starts from zero every session. It doesn't remember why you chose that architecture, what you tried and rejected, or what's left to do.

Engram fixes that. It's a memory system that lives in your git repo (refs/engram/) — tasks, decisions, reasoning, and context, all saved so your agent picks up exactly where it left off.

Integrations:

curl -fsSL https://github.com/vincents-ai/engram/releases/latest/download/install.sh | bash
engram setup workspace

Read the docs | User guide | For agents

Secure, extensible plugin runtime for autonomous agents and microservices. FFI ABI v2, hot reload, OpenTelemetry tracing, Ed25519 signatures.

LLM-powered shell built on shrs with three execution modes (kids/agent/admin). ACL engine, command snapshots, undo/redo, Linux namespace sandboxing. Embeddable as a library for agentic-loop's tool system.

Standardised Nix modules for vincents-ai projects. Reproducible builds, dev shells, deployment across the ecosystem.

We build tools that make EU regulatory compliance (GDPR, DORA, NIS2, EU Cyber Resilience Act) part of your development workflow — not an afterthought.

All projects use trait-first, Git-native design with EU compliance built in.