Commit Graph

6 Commits

Author SHA1 Message Date
3a8a6e98b1 fix(cost): per-model pricing so pi/Paperclip report real spend
The omni provider's models.json entries carried no 'cost' field, so pi's
calculateCost() returned 0 for every run and Paperclip cost events/budgets
all read $0.00 despite real spend.

- scripts/omni-pricing.mjs: pulls authoritative prices (OmniRoute /api/pricing
  for deepseek/*, OpenRouter /api/v1/models for openrouter/*) and injects
  { cost: { input, output, cacheRead, cacheWrite } } USD per 1M tokens.
- Priced: deepseek-v4-flash, deepseek-v4-pro, gemini-3.1-flash-lite, gemini-2.5-flash.
- Must be re-run after every /omni sync, which rewrites models.json wholesale.
2026-09-27 18:37:24 +10:00
3f36f599ce chore(models.json): refresh from /omni sync (476 omni models) 2026-09-27 18:37:14 +10:00
d475607011 Add dashboard.ts extension (pi dashboard state collection) + sync models.json with OmniRoute model list 2026-07-31 14:25:00 +10:00
4085add586 lock provider routing: use only [...] for cache continuity on deepseek, kimi, minimax, qwen 2026-06-14 14:00:22 +10:00
6c2f36b69d fix provider routing: use correct slugs (baidu, streamlake, moonshotai, minimax, qwen) 2026-06-14 08:53:09 +10:00
aaa21a4f2f add 11 new sub-agent types, fix model routing, add models.json
New agents (11):
- chat-search, code-analysis, code-ingest, database
- document-writer, home-automation, image-maker
- iot-coder, iot-hardware, video-analyze, vscode-setup

Fixes:
- Remove broken :deepseek suffix from coder-basic, coder-pro,
  devops-basic, devops-pro (caused silent fallback to parent model)
- Add models.json with openRouterRouting for DeepSeek, Qwen,
  MiniMax, Moonshot AI provider cache targeting

Additions:
- headroom-bridge extension (compress_for_agent tool)
- models.json with provider routing config

Cleanup:
- Remove old pi-subagents extension (replaced by @tintinweb/pi-subagents)
2026-06-13 12:14:17 +10:00