{{ message }}
feat(examples): codex_runtime_on_agentkit — deploy a runtime=codex agent to AgentKit#608
Merged
Merged
Conversation
A minimal deployable app whose agent runs on the OpenAI Codex runtime (`Agent(runtime="codex")`), deployed to Volcengine AgentKit via `veadk agentkit launch`. - agents/codex_agent: Agent(runtime="codex") - app.py: ADK agent API server (deploy entry, + /ping) - scripts/install_veadk.sh: installs veadk from main + openai-codex (not a veadk dep; pulls openai-codex-cli-bin, the Codex binary as a manylinux wheel, so no separate binary install in the Linux build) - README.md / README.zh.md: codex-on-AgentKit specifics, deploy steps, caveats The model in MODEL_AGENT_* is bridged onto Codex's Responses API by the in-process shim, so a normal Ark chat model works unchanged.
…cript veadk-python>=0.5.39 (with the codex runtime) is on PyPI, so the example no longer needs the shallow github clone build script — it installs everything via the default `uv pip install -r requirements.txt`. This is simpler and avoids the flaky build-network → github connectivity that the clone depended on. - requirements.txt: veadk-python>=0.5.39 + openai-codex (+ openai-codex-cli-bin), the latter two pinned to exact pre-release versions so uv installs them without a global --prerelease flag - remove scripts/install_veadk.sh (no longer needed) - README zh/en: document the PyPI install + pre-release pin
zakahan
approved these changes
Jun 15, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A minimal, deployable example: a VeADK agent running on the OpenAI Codex runtime (
Agent(runtime="codex")), deployed to Volcengine AgentKit withveadk agentkit launch.What's inside
How it works
Agent(runtime="codex")— the Runner owns session/memory/tracing; Codex drives the turn. TheMODEL_AGENT_*chat model (Ark) is bridged onto Codex's Responses API by the in-process shim, so a normal Ark chat model works unchanged.uv pip install -r requirements.txt—veadk-python>=0.5.39ships the codex runtime;openai-codex(+openai-codex-cli-bin, the Codex binary as a manylinux wheel) is listed explicitly and pinned to exact pre-release versions souvinstalls the betas without a global--prereleaseflag. No build script / git clone.Verified (real deploy)
Deployed to AgentKit and exercised end-to-end:
agentkit invoke "你好，你叫什么"→ agent replied (reasoning + final text both forwarded — the full codex trajectory).agentkit invoke "用 ls 列出当前目录…再 cat requirements.txt"→ the agent ran the tools inside the runtime container and returned the real file listing + the verbatimrequirements.txtcontents.So the codex tool sandbox (multi-step tool loop) works inside the AgentKit container.
Notes documented in the README: model is bridged (need not be an OpenAI model); first-request latency from the Codex binary spawn; the build install can take a few minutes (re-run reuses cached layers); tool-heavy agents may need runtime permissions.