Convilyn SDK (author)

Build tool servers for the Convilyn AI workflow platform.

Companion package: if you only want to call the Convilyn API (no server hosting), install convilyn instead. The two packages are intentionally separate so consumers don't pay the uvicorn dependency cost (convilyn-author wraps a lightweight internal MCP-over-uvicorn runtime, not FastAPI).

Free to install, paid to deploy. pip install convilyn-author is free. You host your tool server on your own infrastructure (Lambda, Fargate, VM) at your own cost; Convilyn bills the caller of your tools, not you (see DEPLOYMENT §8).

Quick Start

Step 0 — get a developer key (cvl_)

The author track authenticates with a developer key (cvl_ prefix) — your consumer key (ck_, from the web console) will NOT work here. Mint one by registering as a developer; there is no console UI for this yet:

import asyncio
from convilyn_author import ConvilynClient

async def main() -> None:
    result = await ConvilynClient().register(
        email="you@example.com",
        name="Your Name",
        company="Optional Co.",
    )
    print(result["api_key"])   # cvl_… — shown ONCE, store it securely

asyncio.run(main())

Export it as CONVILYN_API_KEY (see Environment Variables below — the client reads CONVILYN_API_KEY, not CONVILYN_DEVELOPER_KEY). A 503 REGISTRATION_STORE_UNAVAILABLE response means the platform side is temporarily unavailable — retry later; it is not a problem with your request.

Step 1 — scaffold a tool server

pip install convilyn-author
convilyn-author init my-server
cd my-server

Workflow authoring lives in the Convilyn chat Builder (web console) — the canonical authoring surface. This is the tool-server SDK: build, register, verify and operate the MCP tool servers the platform's workflows call. The WorkflowSpec DSL and submit_workflow were removed in 2.3.0b1; edge integrators export a Builder-built workflow's grounded contract via the consumer SDK (client.user_workflows.grounded_contract(...)).

Edit server.py:

from convilyn_author import ConvilynServer

server = ConvilynServer(
    name="my-server",
    description="My custom tool server",
    version="1.0.0",
)

@server.tool(description="Process text input")
async def process(text: str) -> dict:
    return {"result": text.upper()}

if __name__ == "__main__":
    # Production entry point: requires CONVILYN_HMAC_SECRET (fail-closed).
    # For local development run `convilyn-author dev` instead — it serves
    # without a secret. Or pass server.run(dev=True) / set
    # CONVILYN_DEV_INSECURE=1 for a bare `python server.py` local run.
    server.run()

CLI Commands

Testing

from convilyn_author.testing import ConvilynTestRunner

runner = ConvilynTestRunner(server=server)
result = await runner.call_tool("process", {"text": "hello"})
assert result.success

report = await runner.run_compliance_check()
assert report.all_passed

Environment Variables

Migrating from convilyn to convilyn-author

The legacy convilyn binary was removed in v2.0.0 — use convilyn-author. The only change is the binary name; all sub-commands, flags, and behaviour are identical:

# Before
convilyn init my-server
convilyn dev --port 8080
convilyn push --endpoint-url https://my-server.example.com

# After
convilyn-author init my-server
convilyn-author dev --port 8080
convilyn-author push --endpoint-url https://my-server.example.com

Why the rename: the consumer SDK (PyPI convilyn, the client for API callers) also ships a convilyn binary. Two packages cannot coexist in one environment while both claiming the same console script. The author SDK now owns convilyn-author; the consumer SDK keeps convilyn.

Next

• DEPLOYMENT.md — Lambda / Fargate / VM walkthroughs + HMAC contract
• STABILITY.md — what the public API is and the SemVer / deprecation promise behind it
• CHANGELOG.md — version history
• examples/ — runnable example tool servers + workflows
• AGENT.md — guidance for AI coding agents contributing to this SDK

Licence

Apache-2.0. See LICENSE.