Native Shell
Native Shell:AI 生成的 UI,在你的應用中原生呈現,涵蓋每個平台。iOS、Android、macOS、Windows 與 Linux 上的應用、GPUI 應用與網頁在執行時載入已簽名、沙箱隔離的 TSX 模組並繪製 th
AI 翻譯
AVP - Command your agent fleet with Open Source
翻譯準備中
取自儲存庫 README
正在載入完整 README。
檢視完整 README ↗Status: Draft v0.1
AVP is an open standard for running and observing AI agents. Shared task and event formats let you run, compare, and inspect agents across different runtimes.
A supervisor sends one JSON Commission with the task, model, and resources. The agent streams a Trajectory of model responses, tool calls, usage, cost, and outcome.
Supervisor ─── Commission ──▶ Agent
Supervisor ◀── Trajectory ─── Agent
See the spec overview for the three contracts and how they fit together.
Install the CLI. Requires uv and a running container engine with a Docker API, such as Docker or podman:
uv tool install https://github.com/portofcontext/agent-voyager-project/releases/download/cli-v0.7.0/avp_cli-0.7.0-py3-none-any.whl
avp hello-world
avp hello-world runs an example eval on three agents. LLM calls are mocked,
so it needs no API key and costs nothing. avp eval share publishes it to a
public URL.
Install a few agents and save a provider key in the vault. The vault command prompts for your key:
avp agent install pydantic-ai
avp agent install claude-code
avp agent install pi
avp vault set anthropic
Save this as task.json:
{
"avp_version": "0.1",
"model": {
"provider": "anthropic",
"id": "claude-haiku-4-5",
"credential": {"proxy": "anthropic"}
},
"prompt": "Say hello."
}
Run the same Commission on each runtime:
avp run --agent pydantic-ai --commission task.json
avp run --agent claude-code --commission task.json
avp run --agent pi --commission task.json
Each run happens in an OpenSandbox sandbox, and the CLI prints the Trajectory's location. See the command guide for credentials, networking and volumes, and the eval guide for comparisons.
List the built-in agents and install one by name:
avp agent list --builtin
avp agent install codex
| Runtime | Install name | Adapter |
|---|---|---|
| Claude Agent SDK | claude-code | avp-claude-agent-sdk |
| OpenAI Agents SDK | openai-agents | avp-openai-agents |
| Codex | codex | avp-codex |
| Pi | pi | avp-pi |
| Gemini CLI | gemini-cli | avp-gemini-cli |
| LangChain Deep Agents | deep-agents | avp-deep-agents |
| Pydantic AI | pydantic-ai | avp-pydantic-ai |
| OpenCode | opencode | avp-opencode |
To integrate another runtime, start with the SDK guide, integration patterns, and an adapter above.
SDKs: Python and TypeScript.
Specs: Commission,
Agent Descriptor, and
Trajectory.
The CLI's avp conformance checks agents against the spec's rules.
Each SDK, CLI, and adapter is an independent project with its own lockfile. Install uv and Node.js 24, then run these commands from the repository root:
make sync
make check
make check makes no paid model calls; CLI sandbox tests run when a container
engine is available.
See the contributor guide, AVP skill, or make help.
Built by Port of Context.