CODEX 0.115.0 SHIPS SUBAGENTS GA, FILESYSTEM RPCS + PYTHON SDK, AND REALTIME TRANSCRIPTION
OpenAI Codex 0.115.0 lands with subagents GA, new app-server filesystem APIs with a Python SDK, and a realtime transcription mode. The release adds full‑resolu...
OpenAI Codex 0.115.0 lands with subagents GA, new app-server filesystem APIs with a Python SDK, and a realtime transcription mode.
The release adds full‑resolution image inspection, a dedicated realtime transcription mode with unified session config, and an app‑server exposing filesystem RPCs for reads/writes, copies, directory ops, and path watching—plus a new Python SDK to use those APIs (release notes, SDK thread). Subagents are now broadly usable, with defaults and custom agents via TOML in ~/.codex/agents, letting you route parallel “explorer/worker” flows or bind agents to faster models overview.
Separately, some developers proxy Codex via ChatGPT OAuth to a local OpenAI‑compatible endpoint, but community discussion calls this a gray area that may be blocked later (repo, HN). A report on free Pro/Codex access for OSS maintainers is circulating, but it isn’t an official source yet summary.
Filesystem RPCs plus a Python SDK let you build safer, auditable agent workflows that read/write project files without raw shell access.
Subagents and a realtime transcription mode enable faster, parallel task orchestration and voice-driven dev loops.
-
terminal
Run an agent pipeline that splits parse/transform/validate into subagents, then measure wall‑clock speedup and failure isolation versus a single agent.
-
terminal
Use the Python SDK to gate all file writes through app‑server RPCs and log changes; compare auditability and rollback to your current approach.
Legacy codebase integration strategies...
- 01.
Wrap existing CLI/script automations with app‑server filesystem RPCs to reduce shell escape risk while keeping your repos and CI structure.
- 02.
Validate subagent inheritance of sandbox/network rules in your environment; the release claims fixes, but confirm against your policy stack.
Fresh architecture paradigms...
- 01.
Design agentic services around the app‑server + Python SDK from day one to centralize IO, approvals, and auditing.
- 02.
Adopt custom TOML agents for distinct roles (explorer/worker) and bind faster models for bursty parallel work.