a1identity Python SDK launches on PyPI

The Python package a1identity has been published to PyPI, introducing a SDK that gives AI agents a verifiable “passport” and a receipt for every authorized action [2]. The library, version 0.1 (initial release), requires Python 3.9+ and a running A1 gateway, which can be started via Docker compose. Once the gateway is live, developers can install the base client with pip install a1 and add framework‑specific integrations such as LangChain, LangGraph, LlamaIndex, AutoGen, CrewAI, Semantic Kernel or OpenAI Agents SDK [2].

The core idea is to close the “Recursive Delegation Gap”: in multi‑agent workflows it is currently impossible to prove which human authorized the final action. A1 addresses this by issuing cryptographic passports and chaining delegation certificates, allowing any downstream agent to present a verifiable receipt [2]. The SDK provides decorators (e.g., @a1_guard) and higher‑level tools that call the gateway’s /v1/passport/authorize endpoint, automatically raising an A1AuthorizationError if the request is not permitted. Sample code shows how to guard a trade function, issue a passport via the a1-cli, and query the gateway directly with an A1Client instance [2].

Beyond the core API, the package ships with optional integrations for key‑management services—AWS KMS, GCP Cloud KMS, HashiCorp Vault Transit, and Azure Key Vault—so that private keys can be stored securely [2]. It also offers SIEM exporters for Datadog, Splunk and OpenTelemetry, enabling audit logs to be streamed to existing monitoring pipelines. A built‑in Model Context Protocol (MCP) server lets compatible agents or IDEs (e.g., Claude Code, Cursor) authorize actions without code changes, simply by adding an entry to an .mcp.json configuration [2].

The release is accompanied by a lightweight web studio at http://localhost:8080/studio, where users can issue passports, inspect delegation chains and manage agents without writing CLI commands. All artifacts are published via trusted publishing using twine and include Sigstore attestation bundles, providing reproducible provenance for both the source tarball and the wheel [1].

If the SDK gains traction, it could become a de‑facto standard for tracing AI‑driven decisions across complex delegation trees, especially as enterprises adopt multi‑agent pipelines. The open question is whether the broader Python AI ecosystem will adopt A1’s model before official support from major cloud providers emerges.