Documents the thinkingLevelMap for the juggernaut and turbo models in the ~/.pi/agent/models.json example, and lowers maxTokens from 128000 to 32000 to match the working configuration. LLM-Generated: true
1.6 KiB
1.6 KiB
Pi AI Harness
Prereqs
I would recommend using pass to store your api key. This is installed with dnf install pass on Fedora systems.
After installing pass, update your ~/.gnupg/gpg-agent.conf to set the timeout to max so automated processes can use the values in your password store:
default-cache-ttl 2147483647
max-cache-ttl 2147483647
Then add a pass.sh to you ~/.bashrc.d that looks like
pass unlock
That way every time you log in your pass keyring is unlocked for your session so automations can use it.
Setup
This is my ~/.pi/agent/models.json
{
"providers": {
"reeseapps-aipi": {
"baseUrl": "https://aipi.reeseapps.com/v1",
"api": "openai-completions",
"apiKey": "!pass aipi/dev-token",
"models": [
{
"id": "juggernaut",
"input": ["text", "image"],
"reasoning": true,
"contextWindow": 262144,
"maxTokens": 32000,
"thinkingLevelMap": {
"off": "none",
"minimal": null,
"low": "low",
"medium": "medium",
"high": null,
"xhigh": "xhigh",
"max": null
}
},
{
"id": "turbo",
"input": ["text", "image"],
"reasoning": true,
"contextWindow": 262144,
"maxTokens": 32000,
"thinkingLevelMap": {
"off": "none",
"minimal": null,
"low": "low",
"medium": "medium",
"high": null,
"xhigh": "xhigh",
"max": null
}
}
]
}
}
}