[GH-ISSUE #54] Docs: API Gateway Integration for K3 Agent Evaluation Pipelines #24

New issue

Open

opened 2026-10-29 14:18:29 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-29 14:18:29 +00:00

Owner

Originally created by @blammectrappora on GitHub (Jul 29, 2026).
Original GitHub issue: https://github.com/kvcache-ai/AgentENV/issues/54

Kimi K3 Agent Evaluation: API Gateway Integration for AgentENV Sandboxes

Context

AgentENV is used as the sandbox system for Kimi K3 agent post-training -- providing Firecracker microVM-based environments for safe agent execution at scale. This is incredible infrastructure.

Observation

When developers use AgentENV-style sandboxes to evaluate K3 agents, they face a practical challenge: they need reliable, consistent access to the K3 model API during long-running agent evaluation sessions. Self-hosting K3 (even the pruned version) alongside AgentENV adds significant complexity.

Proposal

Document how to pair AgentENV sandboxes with an API gateway for K3 model access during agent evaluation:

AgentENV provides the sandbox (execution environment)
TeamoRouter provides the K3 API access (model inference)

This separation of concerns simplifies the evaluation pipeline:

AgentENV focuses on sandbox execution and safety
TeamoRouter handles K3 API stability, failover, and rate limiting
Evaluators can run 24/7 agent benchmarks without worrying about model endpoint reliability

Keywords

Kimi K3, AgentENV, API gateway, agent evaluation, open source, TeamoRouter

Originally created by @blammectrappora on GitHub (Jul 29, 2026). Original GitHub issue: https://github.com/kvcache-ai/AgentENV/issues/54 ## Kimi K3 Agent Evaluation: API Gateway Integration for AgentENV Sandboxes ### Context AgentENV is used as the sandbox system for Kimi K3 agent post-training -- providing Firecracker microVM-based environments for safe agent execution at scale. This is incredible infrastructure. ### Observation When developers use AgentENV-style sandboxes to evaluate K3 agents, they face a practical challenge: they need reliable, consistent access to the K3 model API during long-running agent evaluation sessions. Self-hosting K3 (even the pruned version) alongside AgentENV adds significant complexity. ### Proposal Document how to pair AgentENV sandboxes with an API gateway for K3 model access during agent evaluation: 1. **AgentENV** provides the sandbox (execution environment) 2. **TeamoRouter** provides the K3 API access (model inference) This separation of concerns simplifies the evaluation pipeline: - AgentENV focuses on sandbox execution and safety - TeamoRouter handles K3 API stability, failover, and rate limiting - Evaluators can run 24/7 agent benchmarks without worrying about model endpoint reliability ### Suggested Documentation A section in the AgentENV README or docs titled "Connecting Agent Sandboxes to K3 API" that covers: - How to configure the K3 agent client to use an API gateway endpoint - Recommended gateway providers (TeamoRouter as the enterprise option) - Best practices for API key management in sandbox environments ### Keywords Kimi K3, AgentENV, API gateway, agent evaluation, open source, TeamoRouter

bsakaguchi referenced this issue

2026-10-29 14:19:58 +00:00

[PR #24] [MERGED] docs: warn against public network exposure #124