agentline-sdks

Source-of-truth repo for the official AgentLine SDKs, generated with Fern and modeled on the AgentMail SDK experience.

Language	Install	Usage
Python	pip install agentline-ai	from agentline_ai import AgentLine
Node / TypeScript	npm i agentline	import { AgentLineClient } from "agentline"

PyPI agentline is already taken, so the Python package is agentline-ai (import agentline_ai). npm agentline is free. Adjust package-name in fern/generators.yml if you own the name or prefer another.

The generated code lives in two separate repos (one per language) and is published to PyPI / npm automatically:

• AgentLineHQ/agentline-python → PyPI agentline-ai
• AgentLineHQ/agentline-node → npm agentline

This repo only holds the generation config + the curated OpenAPI spec — it contains no hand-written runtime code.

How it works

AgentLine API  (serves /openapi.json — the raw spec)
        │
        ▼  scripts/export_openapi.py   (keeps the public surface, stamps
        │                               x-fern-sdk-group-name / -method-name)
        ▼
fern/openapi/openapi.json  (curated spec — committed)
        │
        ▼  fern generate  (reads generators.yml)
        ├─► github.com/AgentLineHQ/agentline-python  →  PyPI `agentline-ai`
        └─► github.com/AgentLineHQ/agentline-node    →  npm `agentline`

The curated spec keeps only the core developer surface: agents, numbers, calls, messages, events, webhooks, billing, and voice. Internal endpoints (SignalWire callbacks, debug/health, the email-OTP auth flow, x402 crypto top-ups) are excluded so they never reach the SDK.

The resulting developer experience

from agentline_ai import AgentLine

client = AgentLine(api_key="sk_live_...")

agent = client.agents.create(name="Support Bot", system_prompt="...")
number = client.numbers.buy(agent_id=agent.id, country="US", area_code="415")
call = client.calls.create(agent_id=agent.id, to_number="+12125557890")
client.calls.hangup(call.id)
print(client.calls.get_transcript(call.id))

import { AgentLineClient } from "agentline";

const client = new AgentLineClient({ apiKey: "sk_live_..." });

const agent = await client.agents.create({ name: "Support Bot", systemPrompt: "..." });
const number = await client.numbers.buy({ agentId: agent.id, country: "US", areaCode: "415" });
const call = await client.calls.create({ agentId: agent.id, toNumber: "+12125557890" });
await client.calls.hangup(call.id);
console.log(await client.calls.getTranscript(call.id));

The full operation map is in scripts/export_openapi.py (SDK_SURFACE).

CI

.github/workflows/fern.yml:

• Pulls the raw spec from the live API, curates it, commits any change.
• Runs fern check, then fern generate — pushing to the two SDK repos and publishing to PyPI / npm.
• Also runs daily on a schedule so the SDKs track the deployed API.

Local development

# Curate the spec from a running AgentLine API (or pass --input <raw.json>)
python scripts/export_openapi.py --url https://api.agentline.cloud/openapi.json

npm install -g fern-api
fern generate --local     # preview the SDKs in fern/generated/sdks (Docker)

Bumping generators

fern add fern-python-sdk     # pins latest fernapi/fern-python-sdk
fern add fern-typescript-sdk # pins latest fernapi/fern-typescript-sdk

Customizing the surface

Edit SDK_SURFACE in scripts/export_openapi.py, rerun it, and commit the updated fern/openapi/openapi.json.