Skip to content

User-Defined Models

Early Access

Strikes is currently in early access with trusted partners. Contact us for early access.

Strikes supports standard inference models for quick testing, but you'll likely want to deploy agents using your preferred inference providers, endpoints, and API keys. This is known as User Defined Models. The CLI provides a set of commands under dreadnode models to create and manage local configurations. To use these models in Strike runs, pass -m/--model <id> to dreadnode agent deploy. The configuration will be sent to the server, used for the duration of the run, and discarded afterward.

The Strikes subsystem proxies inference traffic using our Rigging library, specifically with the LiteLLM generator. We apply the same "generator id" concept from Rigging to configure inference on the Strike dropship. Typically, this involves identifying the appropriate generator id for your model in Rigging and then installing it using the CLI.

Example#

Suppose you have a Deepseek API key and want to use one of their models in my agent. After reviewing the LiteLLM docs and the Deepseek model list, you want to make both deepseek-chat and deepseek-reasoner available.

  1. Start by installing the models with the following commands:

    $ dreadnode model add --id deepseek-r1 -n "Deepseek R1" -g deepseek/deepseek-reasoner -k '$DEEPSEEK_API_KEY'

🔧 Added model deepseek-r1 in /Users/nick/.dreadnode/models.yml


$ dreadnode model add --id deepseek-v3 -n "Deepseek V3" -g deepseek/deepseek-chat -k '$DEEPSEEK_API_KEY'

🔧 Added model deepseek-v3 in /Users/nick/.dreadnode/models.yml

$ dreadnode model list

╭─────────────┬─────────────┬──────────┬────────────────────────────┬───────────────────╮
│ ID          │ Name        │ Provider │ Generator ID               │ API Key           │
├─────────────┼─────────────┼──────────┼────────────────────────────┼───────────────────┤
│ deepseek-r1 │ Deepseek R1 │ -        │ deepseek/deepseek-reasoner │ $DEEPSEEK_API_KEY │
│ deepseek-v3 │ Deepseek V3 │ -        │ deepseek/deepseek-chat     │ $DEEPSEEK_API_KEY │
╰─────────────┴─────────────┴──────────┴────────────────────────────┴───────────────────╯
    Wrapping the -k with avoids resolving the API key right now, and tells the CLI to pass it from my ENV anytime I use one of these models.

  2. Now you can pass these models directly to your agent when you deploy it:

    $ dreadnode agent deploy -m deepseek-r1

╭─ run f7eb54b0-0cda-41cf-a15e-d3991be90531 ──────────────────────────────────────────────╮
│                                                                                         │
│   status     completed                                                                  │
│   strike     Simple (simple)                                                            │
│   type       network-ctf                                                                │
│                                                                                         │
│   model      deepseek-r1                                                                │
│   agent      demo [chocolate-condor] (rev 1)                                            │
│   image      dev-registry.dreadnode.io/monoxgas/agents/demo:c39f7c1a                    │
│   notes      -                                                                          │
│                                                                                         │
│   duration   57.8s                                                                      │
│   start      Tue Jan 21 17:47:01 2025                                                   │
│   end        Tue Jan 21 17:47:59 2025                                                   │
│                                                                                         │
│                                                                                         │
│   zone       status      duration   inferences   outputs   score                        │
│  ────────────────────────────────────────────────────────────────                       │
│   web        completed   31.0s          1           1        0                          │
│   database   completed   43.0s          1           2        0                          │
│                                                                                         │
╰─────────────────────────────────────────────────────────────────────────────────────────╯

CLI storage

The CLI stores all user-defined models in ~/.dreadnode/models.yml, which you can manually edit and share.