Future AGI, an open‑source platform released under the Apache 2.0 license, aims to give developers tools for tracing, evaluating, simulating and guardrailing large‑language‑model agents that run on self‑hosted infrastructure.
How the system registers and reports
When a new instance of Future AGI boots for the first time, it automatically registers with the central service. The registration packet includes an instance identifier, the software version, the deployment type and the email addresses and domains of any active administrators. This exchange happens before anyone can log into the dashboard.
Operators can suppress the registration by setting the environment variable FUTURE_AGI_TELEMETRY_DISABLED=1 in the .env file before the initial start. If the variable is added after the first boot, the admin list has already been transmitted and cannot be retracted.
Even with telemetry disabled, a single “census ping” still reaches the service, containing only the instance ID, version and deployment type; email data is omitted. Ongoing heartbeats cease, and the telemetry settings page advises users to “turn networking off at the edge if you need full silence.”
Components and data flow
The installer provisions several back‑end services: ClickHouse, PostgreSQL, Redis, RabbitMQ and Temporal. These sit behind a local dashboard reachable at localhost:3000. All prompt text, model output and tool calls are routed through a tracer before being stored in ClickHouse.
Instrumentation supports more than fifty agent frameworks, among them LangChain, LlamaIndex, CrewAI and DSPy, using OpenTelemetry standards. Access controls applied to the database also govern the trace data, giving defenders a clear line of sight into agent activity.
The platform’s gateway, called the Agent Command Center, presents an OpenAI‑compatible proxy that can interface with over a hundred providers. It offers routing strategies, semantic caching, virtual keys, multi‑cloud provisioning and agent‑to‑agent communication. All provider credentials terminate at this gateway, limiting exposure.
Built‑in security scanners—18 in total—cover personal‑identifiable information leaks, jailbreak attempts and prompt‑injection attacks. Vendor adapters for Lakera, Presidio and Llama Guard are available both inline within the gateway and as separate SDK components. Inline scanning keeps the 99th‑percentile latency at or under 21 ms according to the project’s benchmark suite.
The codebase is hosted publicly on GitHub, where anyone can clone, modify or contribute to the project.
Documentation notes that both air‑gapped and on‑premise deployments are supported, and that the platform operates without a mandatory phone‑home requirement.
Related: Moonshot Unveils Kimi K3 Weights to Select Few
In practice, this means organizations that already run self‑hosted LLM workloads can plug Future AGI into existing pipelines without exposing internal data to external services.
The combination of tracer‑driven storage, granular access controls and low‑latency scanning offers a practical path to monitor AI agents that evolve over time.
Future AGI’s approach reflects a broader trend toward giving enterprises more visibility into autonomous AI systems. By keeping core components on‑premise and limiting outbound traffic, the platform reduces the attack surface that typically accompanies cloud‑based AI services.
For teams that must comply with strict data‑handling policies, the ability to run all telemetry and security checks locally can be a decisive factor.
Availability and next steps
Developers interested in evaluating the platform can download it for free from the public repository. The installation script automates the setup of the required services and launches the dashboard.
Ongoing development is tracked on the same repository, where contributors can submit issues or pull requests.
While the platform promises extensive coverage of agent frameworks, users should verify compatibility with any custom integrations they maintain.
The documentation advises reviewing the list of supported providers and testing the gateway’s routing logic before deploying to production environments.
Future AGI is available for free on GitHub.
