AI Agent Stack for Berlin and Germany's Startup Capital

A startup in the capital often already pays a model lab. The research feed, the agent that holds a German and English brief, and a machine for the evaluation script are still separate invoices. SandBase puts them on one account:

• 2,000+ models and data APIs from 100+ platforms, drawn on a single balance
• Stored Agents and open Sessions, in the OpenAI Agents shape and in OpenAgentCore
• An E2B sandbox when a job must open a file or run a command
• SKILL.md so the coding assistant can read the catalog

An Agent or a sandbox can start with no charge. Qualifying data APIs share 100 free calls each day inside the organization. The allowance resets at 00:00 UTC. Model calls are priced on their own listing.

Use https://api.sandbase.ai/v1 and the header Authorization: Bearer $SANDBASE_API_KEY.

These public pages describe the city. SandBase is not announcing a Berlin address.

• In the GSER 2026 Top 40, Berlin is #22 globally and #3 in Europe. The same report lists an ecosystem value of $89.5 billion for H2 2023–2025. Startup Genome’s Berlin profile describes a startup scene that blends creativity with tech innovation.
• Berlin Partner says Berlin is already Germany’s top location for AI: more than 30% of the country’s 687 AI startups are in the capital. The same note names two national competence centers in the city, DFKI and BIFOLD, and says companies offering their own AI services generate €2.4 billion in revenue.
• TU Berlin calls itself one of the leading places in Germany for research and teaching in artificial intelligence and computer science. It describes BIFOLD as one of Germany’s six national AI competence centers, focused on machine learning and big data, with permanent funding from the federal government and the State of Berlin.

Künstliche Intelligenz in Berlin sits on that mix of startups and labs. SandBase is the API under it: one key, rather than a new vendor for every model or dataset.

1. Create the organization key in the Console.
2. Define the Agent with the OpenAI SDK or POST /v1/agents: a model, instructions, and tools.
3. When projection_status is ready, open a Session and send the task.
4. If the task needs files or a script, use the E2B SDK at https://sandbox.sandbase.ai.
5. Or pass SKILL.md to the assistant and let it look up the catalog.

Longer paths: Docs · Agents API · Sandbox quickstart

The OpenAI Python SDK pattern from the SandBase docs, with instructions for Berlin and German readers.

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["SANDBASE_API_KEY"],
    base_url="https://api.sandbase.ai/v1",
)

agent = client.beta.agents.create(
    model="deepseek/deepseek-v4-flash",
    name="Berlin startup brief",
    instructions=(
        "Write for a Berlin and German startup audience. "
        "Compare more than one source and cite each."
    ),
    metadata={"city": "berlin", "team": "germany"},
)

print(agent.id, agent.version)  # agt_..., 1
# Open a Session only after projection_status is ready.

curl -X POST https://api.sandbase.ai/v1/agents/sessions \
  -H "Authorization: Bearer $SANDBASE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"agent_id": "agt_..."}'

Keep scripts and files in a sandbox. Start with the Sandbox quickstart.

If you searched for an AI agent API Berlin teams in Germany can adopt while keeping the OpenAI Agents SDK, SandBase is that API.

Use these public pages. Do not add a SandBase office, a phone number, or a latency figure for Berlin:

• GSER 2026
• Startup Genome — Berlin ecosystem
• Berlin Partner — AI hub
• TU Berlin — AI and information technology

The #22 global rank, the #3 Europe rank, and the ecosystem-value figure are Startup Genome’s for GSER 2026. Berlin Partner is the source for the share of German AI startups and the revenue line. TU Berlin is the source for the BIFOLD description. SandBase does not publish a Berlin phone number, a street address, or a city-specific latency SLA on this page.