pacsys

Pure-Python library for Fermilab's control system.

About

ACNET (Accelerator Control NETwork) is the control system used at Fermilab's particle accelerators. PACSys provides a simple Python interface to interact with ACNET data without needing to understand the underlying protocols.

Features

• Read/Write/Stream any ACNET data types with synchronous or async APIs
• Multiple backends to connect to DPM, DMQ, and ACL
• Full DRF3 parser for data requests with automatic conversion
• Utilities for device database, SSH tunneling, and more
• Low-level ACNET protocol handlers for GETS32/RETDAT/FTPMAN
• Command-line tools like in EPICS - acget, acput, acmonitor, acinfo
• MCP server for AI agent integration (Claude Code, etc.) with supervised writes

Installation

pip install pacsys            # Includes Kerberos and SSH dependencies
pip install "pacsys[all]"     # + Parquet and MCP dependencies

Device API (recommended)

import pacsys
from pacsys import Device, Verify, KerberosAuth

# Create a device -- DRF is validated immediately
dev = Device("M:OUTTMP")

# Read different properties
temperature = dev.read()               # READING (scaled value)
setpoint = dev.setting()               # SETTING property
is_on = dev.status(field="on")         # STATUS field ON
alarm = dev.analog_alarm()             # ANALOG alarm

# Full reading with metadata
reading = dev.get()
print(f"{reading.value} {reading.units}")  # e.g. "72.5 DegF"

# Write with automatic readback verification
result = dev.write(72.5, verify=Verify(tolerance=0.5))
assert result.confirmed

# Control commands with shortcuts
dev.on()
dev.off()
dev.reset()

# Device database metadata (scaling, limits, units)
info = dev.info()
print(info.description)                # "Outside temperature"
print(info.reading.common_units)       # "DegF"
print(info.reading.min_val)            # 0.0

# Block until next reading arrives
reading = dev.with_event("p,1000").await_next(timeout=5)
print(reading.value)

# Stream data
with dev.with_event("p,1000").subscribe() as stream:
    for reading, handle in stream.readings(timeout=10):
        print(reading.value)

# Immutable -- modifications return new instances
periodic_dev = dev.with_event("p,1000")
sliced_dev = dev.with_range(0, 10)

Backend API

import time
import pacsys

# Read a device value through global backend
temperature = pacsys.read("M:OUTTMP")
print(f"Temperature: {temperature}")

# Stream real-time data through global backend
with pacsys.subscribe(["M:OUTTMP@p,1000"]) as stream:
    for reading, handle in stream.readings(timeout=30):
        print(f"{reading.name}: {reading.value}")

# Stream with callback dispatch mode through dedicated DPM instance
# WORKER (default): callbacks on dedicated worker thread, protects event loop
# DIRECT: callbacks inline on reactor thread (lower latency)
with pacsys.dpm(dispatch_mode=pacsys.DispatchMode.DIRECT) as backend:
    handle = backend.subscribe(
        ["M:OUTTMP@p,1000"],
        callback=lambda r, h: print(r.value),
    )
    time.sleep(10)
    handle.stop()

# Write through authenticated DPM instance (requires kerberos ticket)
with pacsys.dpm(auth=pacsys.KerberosAuth(), role="testing") as backend:
    backend.write("Z:ACLTST", 72.5)

Async capabilities

Native async versions with the same API surface.

import pacsys
import pacsys.aio as aio
from pacsys.aio import AsyncDevice

# Module-level API (mirrors pacsys.read, pacsys.get, etc.)
value = await aio.read("M:OUTTMP")
reading = await aio.get("M:OUTTMP")

# Explicit async backend
async with aio.dpm(auth=pacsys.KerberosAuth()) as backend:
    await backend.write("Z:ACLTST", 72.5)

    # Async streaming
    async with await backend.subscribe(["M:OUTTMP@p,1000"]) as stream:
        async for reading, handle in stream.readings(timeout=30):
            print(f"{reading.name}: {reading.value}")

    # AsyncDevice
    dev = AsyncDevice("M:OUTTMP", backend=backend)
    temp = await dev.read()
    await dev.on()

SSH Utilities

Port tunneling, SFTP, and interactive processes over multi-hop SSH.

import pacsys

# Execute commands with automatic Kerberos auth
with pacsys.SSHClient(["jump.fnal.gov", "target.fnal.gov"]) as ssh:
    result = ssh.exec("hostname")
    print(result.stdout) # target

# ACL can be run on the fly - beam switch, DB, etc.
with pacsys.SSHClient("clx01.fnal.gov") as ssh:
    result = ssh.acl("read M:OUTTMP") # "M:OUTTMP       =  72.500 DegF"

MCP Server (AI Agent Integration)

Expose pacsys as an MCP server for AI agents like Claude Code. Read-only by default, granular write policy engine available.

{
  "mcpServers": {
    "pacsys": {
      "command": "python",
      "args": ["-m", "pacsys.mcp"]
    }
  }
}

Experimental Utilities

pacsys.exp provides high-level utilities for experiment workflows:

from pacsys.exp import Monitor, read_fresh, watch, scan, DataLogger, CsvWriter, ParquetWriter

# Collect readings for 10 seconds
result = Monitor(["M:OUTTMP@p,1000", "G:AMANDA@e,8f"]).collect(duration=10)
print(result.mean("M:OUTTMP@p,1000"))
print(result.median("M:OUTTMP@p,1000"))

# Continuous monitoring with blocking wait
mon = Monitor(["M:OUTTMP@p,1000"])
mon.start()
reading = mon.await_next("M:OUTTMP@p,1000")  # block until next reading
snap = mon.snapshot()                          # all buffered data still available
mon.stop()

# Time-slice and export
sliced = result.slice("M:OUTTMP@p,1000", start=t0, end=t1)
timestamps, values = result.to_numpy("M:OUTTMP@p,1000")
df = result.to_dataframe(relative=True)  # elapsed seconds index

# Wait for fresh readings (with multi-count and stats)
results = read_fresh(["M:OUTTMP@p,1000"], count=10, timeout=5.0)
print(results[0].mean())  # windowed stats: mean, std, median, min, max

# Watch for a condition
reading = watch("M:OUTTMP@p,1000", lambda r: r.value > 75, timeout=30)

# Parameter scan with restore
result = scan("Z:ACLTST", ["M:OUTTMP"], values=[0.0, 1.0, 2.0], settle=0.5)

# Log to CSV
with DataLogger(["M:OUTTMP@p,1000"], writer=CsvWriter("log.csv")):
    time.sleep(60)

# Log to Parquet (typed columns, ZSTD compression)
with DataLogger(["M:OUTTMP@p,1000"], writer=ParquetWriter("log.parquet")):
    time.sleep(60)

CLI Tools

EPICS-style command-line tools:

# Read devices
acget M:OUTTMP Z:ACLTST
acget --format json M:OUTTMP

# Write devices (requires authentication: -a kerberos, or -a jwt with -b grpc)
acput -a kerberos Z:ACLTST 72.5
acput -a kerberos -b dmq --verify --tolerance 0.5 Z:ACLTST 72.5

# Monitor (streaming on default event or custom one)
acmonitor M:OUTTMP
acmonitor -n 10 M:OUTTMP@p,500

# Device info (DB + property reads)
acinfo -v M:OUTTMP

Also aliased under pacsys-get, pacsys-put, pacsys-monitor, pacsys-info.

Requirements

• Python 3.10+
• For writes: Kerberos credentials with appropriate role or console class assigned
• For some utilities: must run on the controls network or have SSH access to it

Documentation

See the full documentation for guides, API reference, and protocol details.