Deployment

An agent's definition and governance are the same wherever it runs — choose where based on your data and infrastructure needs.

Mode	Where it runs	Choose it for
Web	Managed aixplain cloud (serverless)	Fastest path to production; no infrastructure to manage; aixplain handles autoscaling
On-prem	Your own infrastructure, including air-gapped	Data sovereignty, regulatory compliance, or zero outbound connectivity
Desktop	Your own machine (aixplain Desktop)	Running locally with local or cloud models, keeping work on-device

---

Web (serverless)

Serverless is the default mode: build an agent in the app or via the SDK, aixplain hosts it, and you invoke it with an API key. There is no provisioning step, autoscaling is handled for you, and runs are monitored in the dashboard.

Prerequisites: an aixplain account, a workspace API key, and the agent's agent_id.

Invoke a deployed agent (Python)

import os, time, requests

API_KEY = os.environ["AIXPLAIN_API_KEY"]
AGENT_ID = os.environ["AIXPLAIN_AGENT_ID"]
headers = {"x-api-key": API_KEY, "Content-Type": "application/json"}

# Start a run
run = requests.post(
    f"https://platform-api.aixplain.com/v2/agents/{AGENT_ID}/run",
    headers=headers,
    json={"query": "Summarize this ticket and suggest next steps."},
    timeout=30,
).json()
request_id = run["requestId"]

# Poll for the result
while True:
    result = requests.get(
        f"https://platform-api.aixplain.com/sdk/agents/{request_id}/result",
        headers=headers,
        timeout=30,
    ).json()
    if result.get("completed"):
        print(result)
        break
    time.sleep(2)

Or with the SDK: aix.Agent.get("agent-id").run(query="..."). Monitor usage, latency, cost, and traces in the dashboard, and capture requestId on every call to correlate runs with your own logs.

Before going live

• Validate behavior on representative and adversarial inputs, and in the app's run traces.
• Add Inspectors for safety, quality, and compliance checks.
• Set access and quotas with API keys and rate limits.
• Keep retries and a primary/secondary fallback chain enabled for model and tool failures, and set clear termination criteria to avoid runaway loops.

---

On-prem

Run the entire aixplain platform inside your own infrastructure, including fully air-gapped environments with zero outbound connectivity — installation, updates, license validation, and all runtime operations are self-contained within your network. aixplain staff have no access to your environment.

• Deployment: containerized services on Linux (Docker / Docker Compose). Scale out behind a load balancer for concurrency and high availability; GPU nodes for model serving are customer-provisioned.
• Air-gapped install: delivered as signed offline bundles; updates are signed patch bundles verified by SHA-256 checksums; perpetual licenses validate locally with no call-home.
• Data handling: all data, models, logs, and telemetry stay within your network. Inference runs in memory and nothing is written to disk by default. Prompts and responses are never used for training. The only opt-in exceptions — RAG embeddings and configured agent or session memory — are governed by RBAC and model-scoped keys.
• Compliance: supports data-residency and access-governance requirements (for example SDAIA / NCA / GCC data protection), enforced entirely within your infrastructure.

Not all services are available on-prem. To scope and set up an on-prem deployment, contact us.

---

Desktop

Run agents locally on your own machine with aixplain Desktop — use local or cloud models and tools, and keep your work on-device. See the Desktop guide.