Editing Openai/694057b6-101c-8007-9a65-c40578c7252d (section)

==== This module provides: ====
* Subprocess supervision for llama-server: - start/stop/restart - hard guarantees against orphan processes
* Async request execution: - submit() returns immediately (with a request id) - caller polls for status/result
* Admission control via concurrency slots: - no internal queue by default
* Robust failure handling: - detect dead/stalled/unreachable server - restart the subprocess - fail in-flight requests with explicit reasons
* Prompting with a worker-owned BIOS system prompt layer: - hivemind/cooperation guidance - runtime metadata (date/time, tool budgets) - stable, testable, separately generated
* Tools: - normal OpenAI-format tool calling with a pluggable ToolRunner (round-trip) - fallback tool-call parsing method when needed
* Exit tools (one-way control signals): - provided at init - recorded when emitted by the model - never alter control flow (beyond whatever the model does)
* Internal streaming: - used for monitoring, progress detection, loop detection - progress = any data flowing after headers
* Loop mitigation: - rely on max_tokens - plus a simple repeated-line early-kill detector