CeaCtl

Cisco Enterprise API Control (ceactl) is a Go CLI for querying Cisco MDS and UCS Manager devices. It currently supports MDS inventory and firmware queries, UCSM server inventory, and an experimental MDS log-analysis workflow with local Ollama models.

Features

• Cisco MDS NX-API integration over HTTPS
• Cisco UCS Manager XML API integration over HTTPS
• MDS firmware version and hardware inventory queries
• UCSM server inventory queries
• MDS log collection or local log-file input
• Mechanical log grouping by severity, facility, mnemonic, interface, and VSAN
• Optional LLM-assisted log analysis through the Ollama Chat API
• Configurable device profiles and verbose request logging

Requirements

• Go 1.26.3 or a compatible version matching go.mod
• Network access and valid credentials for the target MDS or UCSM device
• NX-API enabled on MDS devices
• Ollama and a downloaded model for LLM log analysis (optional)

Build

go mod download
go build -o ceactl .

On Windows, use go build -o ceactl.exe . and replace ./ceactl with ./ceactl.exe in the examples below.

Configuration

Copy the example configuration before running the CLI:

cp .config.yaml.example .config.yaml

devices:
  mds-lab:
    type: mds
    host: 192.0.2.10
    port: "8443"
    username: admin
    password: change-me
    insecure_tls: true

  ucsm-lab:
    type: ucsm
    host: 192.0.2.20
    port: "443"
    username: admin
    password: change-me
    insecure_tls: true

llm_analysis:
  enabled: false
  backend: ollama
  ollama:
    endpoint: http://localhost:11434
    model: gemma4:e2b
  output:
    translate: true
    target_lang: ko_KR

Each device must define type, host, port, username, and password. When exactly one device of the requested type exists, CeaCtl selects it automatically. When multiple devices of that type exist, choose one with --device.

insecure_tls: true disables TLS certificate verification and is intended for lab devices or self-signed certificates. Use certificate verification in production environments. The configuration contains plaintext credentials; .config.yaml is ignored by Git, but it should still be protected with appropriate file permissions.

Set llm_analysis.output.translate: true to send the completed English analysis through a second Ollama request. target_lang selects the translation locale and is required when translation is enabled. Set translate: false to skip the second request.

Commands

Command	Description
ceactl mds version	Show the MDS hostname, firmware version, and uptime.
ceactl mds inventory	Show MDS component names, product IDs, and serial numbers.
ceactl mds logs	Fetch and print locally compressed MDS logs without LLM analysis.
ceactl mds logs analyze	Fetch, group, and optionally analyze MDS logs.
ceactl ucsm servers	Show UCSM server DN, model, serial number, and operational state.

The mds and ucsm command groups share these flags:

Flag	Description
--config <path>	Configuration file path (default: .config.yaml).
--device, -d <name>	Device profile name from the configuration.
--verbose, -v	Print request/session details; log analysis also prints the complete compressed input, cited source evidence, and Ollama API response.

Examples:

./ceactl mds version --device mds-lab
./ceactl mds inventory -d mds-lab
./ceactl ucsm servers --device ucsm-lab
./ceactl mds version --config ./configs/lab.yaml --verbose

Print a locally compressed show logging logfile report without contacting an LLM backend:

./ceactl mds logs --device mds-lab
./ceactl mds logs --file ./samples/mds.log
./ceactl mds logs --file ./samples/mds.log --from 20260701 --to 20260715

The report includes compressed event and sequence rows. Unparsed source lines are limited to the first 10 entries so the command does not print the complete raw logfile. --file reads the local file directly and does not load a device configuration or contact an MDS switch. --from and --to apply the same inclusive YYYYMMDD date filtering used by mds logs analyze.

MDS Log Analysis with Ollama

mds logs analyze parses and semantically compresses the log locally before sending pipe-delimited events to Ollama. The default output contains the model's analysis followed by one-line summaries for individually cited Event IDs such as E1 and E2. Use --verbose to also display the complete compressed input and the exact source messages behind cited events.

When translation is enabled, the CLI sends only the completed analysis text and configured target_lang in a second Ollama request. Output is ordered as the English analysis, cited Event ID summaries or verbose source evidence, and then the translation.

Before writing those final LLM sections, CeaCtl verifies that the translation preserves the sets of Event IDs and protected technical tokens, retains source numeric literals and Markdown heading/list structure, and uses the expected script for supported locales. Repetition and paragraph wrapping may differ between languages. A failed check suppresses the final LLM output and returns an English LLM translation validation failed error.

1. Install and start Ollama

Install Ollama by following the official instructions for Windows, macOS, or Linux. Make sure the server is running; depending on the installation, starting the Ollama app may do this automatically. It can also be started from a terminal:

ollama serve

2. Download the configured model

The model name must be identical in Ollama and .config.yaml:

ollama pull gemma4:e2b
ollama ls

You can use another chat-capable Ollama model by changing both the pull command and llm_analysis.ollama.model.

3. Enable the backend

llm_analysis:
  enabled: true
  backend: ollama
  ollama:
    endpoint: http://localhost:11434
    model: gemma4:e2b

Set endpoint to the Ollama server's base URL without /api; CeaCtl appends /api/chat internally. The current Ollama client does not support API-key or other authentication headers.

4. Analyze device or local logs

Fetch show logging logfile from an MDS device:

./ceactl mds logs analyze --device mds-lab

Analyze an existing local file without contacting the MDS device:

./ceactl mds logs analyze --file ./samples/mds.log

Filter either source by an inclusive date range in YYYYMMDD format:

./ceactl mds logs analyze \
  --device mds-lab \
  --from 20260701 \
  --to 20260715

./ceactl mds logs analyze \
  --file ./samples/mds.log \
  --from 20260701 \
  --to 20260715

--from and --to are optional. A local-file run still requires a valid configuration file with at least one device because the command loads the global configuration before reading the file.

The Ollama request is non-streaming, uses a 128K requested context window and temperature 0, and can wait for up to 10 minutes. Large logs or a model's first load may therefore take some time; the CLI prints an elapsed-time counter while it waits.

Use --verbose to print the complete compressed log report and cited source evidence, plus the complete /api/chat JSON response to stderr. The response includes Ollama's performance fields such as total_duration, load_duration, prompt_eval_count, prompt_eval_duration, eval_count, and eval_duration:

./ceactl mds logs analyze --file ./samples/mds.log --verbose

LLM output is preliminary troubleshooting material and may be incomplete or incorrect. Use --verbose to confirm cited events against the source log details, and collect additional diagnostics before making operational or hardware decisions. If endpoint points to another host, the grouped log content is sent to that host.

Project Structure

.
|-- main.go                         # CLI entry point
|-- cmd/
|   |-- root.go                     # Root command
|   |-- mds/                        # MDS commands and flags
|   `-- ucsm/                       # UCSM commands and flags
|-- internal/
|   |-- config/                     # YAML loading and device selection
|   |-- llm/
|   |   `-- ollama/                 # Shared Ollama Chat API client
|   |-- mds/
|   |   |-- commands/               # MDS query operations
|   |   |-- llmanalysis/            # Analysis prompts and evidence helpers
|   |   |-- llmtranslation/         # Translation-only LLM prompts
|   |   |-- logcompressor/          # Log parsing, grouping, and evidence
|   |   |-- receiver/               # NX-API JSON response parsing
|   |   `-- transceiver/            # NX-API HTTPS client
|   `-- ucsm/
|       |-- commands/               # UCSM query operations
|       |-- receiver/               # UCSM XML response parsing
|       `-- transceiver/            # UCSM XML session and HTTPS client
`-- .config.yaml.example            # Configuration template

Test

go test ./...

Roadmap

• Additional MDS and UCSM commands
• JSON and other output formats
• File output support
• Confirmed MDS configuration changes, such as zones and zonesets
• NX-API and UCSM API-level error parsing
• Relative log windows such as --around and --window
• Configurable LLM output language and translation

License

This project is licensed under the MIT License.