agent-zero-model-sync

A pi extension that keeps the agent-zero-venice provider in ~/.pi/agent/models.json in sync with the models Venice.ai marks as NEW (created within the last 30 days).

• Text-output models only; beta models included.
• Replaces the provider's model list (does not merge).
• Runs automatically at session_start, at most once per local calendar day.
• Reasoning-effort (thinking level) metadata is not trusted from Venice's API — levels are resolved with first-party vendor docs: 0. vendor-efforts.json — curated table verified against first-party docs (platform.kimi.ai, ai.google.dev, platform.claude.com, docs.qwencloud.com, api-docs.deepseek.com)
  1. model_spec.capabilities.reasoningEffortOptions (explicit API options)
  2. "Model support" tables on the official Venice reasoning docs
  3. Description mentions tunable/adjustable/configurable thinking → low/medium/high
  4. supportsReasoningEffort flag alone

Install

pi install npm:agent-zero-model-sync
# or from git:
pi install git:github.com/<you>/[email protected]

One-time setup

The extension only maintains the models array of the provider — you provide the endpoint and key. Add this skeleton to ~/.pi/agent/models.json:

{
  "providers": {
    "agent-zero-venice": {
      "name": "Agent Zero Venice",
      "baseUrl": "https://llm.agent-zero.ai/v1",
      "api": "openai-completions",
      "apiKey": "YOUR-API-KEY",
      "models": []
    }
  }
}

name, baseUrl, api, and apiKey are preserved on every sync; only models is replaced. After a sync lands, restart pi (or run /reload) to pick up the new models.

Commands

| Command | Effect | |---------|--------| | /agent-zero-sync | Force-refresh now (bypasses the daily limit) | | /agent-zero-sync status | Show the last sync result |

Files written

• ~/.pi/agent/models.json — the provider's models array (previous copy kept at models.json.bak-agent-zero-sync)
• ~/.pi/agent/agent-zero-model-sync/state.json — daily-sync marker and last-run summary

PI_CODING_AGENT_DIR is respected for all paths.

Security note

pi extensions run with your full system permissions. This one only reads Venice's public model endpoints and writes the two files listed above — review the source before installing.

Development

npm install                      # types only (pi itself is a peer)
curl -o /tmp/reasoning.html https://docs.venice.ai/guides/features/reasoning-models
node test.mts                    # logic tests + one forced live sync

Re-verify entries in vendor-efforts.json when vendors ship new models. Publish with npm publish.