Future AGI
A self-hosted AI agent evaluation platform with text and voice simulations, OpenTelemetry tracing, and support for Ollama and vLLM.
Future AGI helps teams test AI agents, investigate production failures and compare prompt changes. It brings evaluation and monitoring into the same platform so teams can connect poor responses to the steps that caused them.
- Conversation simulations: Test text and voice agents with multi-turn conversations, personas and adversarial inputs, including voice integrations with LiveKit, VAPI, Retell and Pipecat.
- Response evaluation: Check hallucinations, grounding, tool-use correctness and personal information, or define custom scoring criteria; evaluators and their prompts are inspectable.
- Agent tracing: OpenTelemetry traces show conversation paths, latency and token costs across frameworks such as LangChain, LlamaIndex and CrewAI.
- Prompt optimization: Use production traces as data for prompt experiments and compare candidate prompts against evaluation results.
- Gateway and guardrails: An OpenAI-compatible gateway handles model routing and semantic caching, while guardrail scanners check for prompt injection, jailbreaks and personal information.
The open-source core uses Apache 2.0 and supports Docker self-hosting on a laptop or server. A managed cloud service is also available. Local model backends include Ollama, vLLM and LM Studio; cloud connections include OpenAI, Anthropic and Google Gemini. Self-hosting keeps the platform on your infrastructure, but requests routed to cloud model providers go to those services.
