<> agent.forge

~ agent.forge on main · deploying ideas

We build agents that ship.

A two-track studio out of Munich: battle-tested systems engineering on one rail, applied AI & agent development on the other. We meet in the middle, where shipped software lives.

See what we do → $ send signal
20+ yrs
in production systems
EU-hosted
data stays in Europe
Senior only
no juniors on your code
GMT+1
reply within 1 working day

~/services

Three rails.
One shipping crew.

We don't separate "AI people" from "systems people". The work crosses lanes hourly, and you get the same hands from probe to hand-off.

01
Systems Engineering

Servers that quietly outlive contracts.

Linux infrastructure, observability, deploy pipelines, and the boring runbooks that turn 3 AM pages into well-rested mornings.

  • ▸ Linux ops (Debian/Ubuntu), Docker, Kubernetes when you actually need it
  • ▸ Caddy / Nginx, Postgres, Redis, S3-compatible storage
  • ▸ CI/CD with GitHub Actions, ArgoCD, or whatever your shop runs
  • ▸ SLOs, alerts, dashboards — not just metrics, but answers
02
AI Consulting

Pick the right model. Skip the hype.

Where AI buys you leverage and where it costs more than it earns. Audits, vendor selection, evals, and pragmatic roadmaps you can defend in a budget review.

  • ▸ Use-case discovery & honest ROI sizing
  • ▸ Vendor audits (Anthropic, OpenAI, Mistral, open-weight)
  • ▸ Eval harnesses — measure before you commit
  • ▸ EU data residency & DSGVO-compatible patterns
03
Agent Development

Agents that do, not just chat.

Custom agents with real tool access, durable state, and reliable hand-offs to humans. Built on Claude, GPT, or open weights — wherever it fits.

  • ▸ Tool-using agents (file I/O, web, your APIs, your databases)
  • ▸ MCP servers tailored to your data and workflows
  • ▸ Memory & context strategies that survive long horizons
  • ▸ Human-in-the-loop guardrails and audit trails

~/stack

We speak native in:

Linux Docker Kubernetes Caddy Nginx Postgres Redis Anthropic OpenAI Mistral Llama Ollama vLLM Python TypeScript Rust Go Astro SvelteKit Next.js Terraform Ansible Pulumi MCP LangChain pgvector GitHub Actions Hetzner IONOS Cloudflare Linux Docker Kubernetes Caddy Nginx Postgres Redis Anthropic OpenAI Mistral Llama Ollama vLLM Python TypeScript Rust Go Astro SvelteKit Next.js Terraform Ansible Pulumi MCP LangChain pgvector GitHub Actions Hetzner IONOS Cloudflare Linux Docker Kubernetes Caddy Nginx Postgres Redis Anthropic OpenAI Mistral Llama Ollama vLLM Python TypeScript Rust Go Astro SvelteKit Next.js Terraform Ansible Pulumi MCP LangChain pgvector GitHub Actions Hetzner IONOS Cloudflare

~/approach

A loop, not a launch.

Engagements run in four phases. Each phase has a defined exit — including the option to walk away cleanly.

  1. #01 $ probe

    Probe

    Two-week deep read. Code, infra, runbooks, calendars. We name the constraints out loud before touching anything.

  2. #02 $ pilot

    Pilot

    Smallest demonstrable slice in production-shaped staging. You can kill the project here without sunk cost.

  3. #03 $ productionize

    Productionize

    Hardening, observability, evals, docs. Boring on purpose. We design for the on-call who replaces us.

  4. #04 $ handoff

    Hand-off

    Runbooks, training, four weeks of standby. Then you own it — calmly.

// invariant: at every phase boundary you can kill the project with no further obligation.

~/about

Senior craft, no juniors on your code.

Agent Forge is a deliberately small studio out of Munich, run by senior engineers with two decades of production scars between them.

We don't subcontract your code to a learning loop. The people who scope the work are the people who write it, deploy it, and answer pages about it on Sunday.

Data stays in the EU by default. Conversations happen in English or German. Invoices in Euro. No surprises.

~/contact

Got something non-trivial
to ship?

Send a short brief. We reply within one working day with either an honest "yes, let's talk" or a clear "this isn't us — try X instead".

$ send signal → Book 20 minutes
@ hello@agent-forge.dev ⌖ Munich, Germany ⏱ GMT+1, replies < 24h