polymath-agent 0.1.0 → 0.3.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +31 -2
- package/dist/cli.js +1218 -132
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -93,8 +93,37 @@ poly usage # cost by date + model
|
|
|
93
93
|
| `poly recommend <goal>` | Pre-run recommendation: cheapest / best-value / best-quality model combos + savings. |
|
|
94
94
|
| `poly models` | Browse the catalog with pricing, tier, tool support. Filters: `--tier`, `--tools`, `--search`. |
|
|
95
95
|
| `poly usage` | Recorded usage & cost grouped by **date + model**. `--today`, `--since`, `--sync`. |
|
|
96
|
-
| `poly
|
|
97
|
-
| `poly
|
|
96
|
+
| `poly analyze` | **Which approach reaches the goal with the fewest tokens** — efficiency playbook, best model per task type, objective × achievement, usage per command. |
|
|
97
|
+
| `poly sync` | Push **distilled efficiency insights** to Firebase ([Data Connect SQL](dataconnect/) / Firestore). Raw logs stay local unless `--raw`. |
|
|
98
|
+
| `poly config show\|set\|firestore\|dataconnect\|local` | View/change settings. |
|
|
99
|
+
|
|
100
|
+
After each `poly run`, rate the result 0–9 (one keypress) — your goal-achievement
|
|
101
|
+
rating joins the auto score (completed/planned steps) to power `poly analyze`.
|
|
102
|
+
|
|
103
|
+
### The efficiency playbook (learned routing)
|
|
104
|
+
|
|
105
|
+
Everything is captured locally (SQLite). `poly analyze` distills it into a **playbook**
|
|
106
|
+
of *notably* efficient approaches — a (task, model) pair qualifies only with ≥3
|
|
107
|
+
successful runs, ≥70% success, and **≥20% fewer tokens than the median** of its
|
|
108
|
+
competitors. The playbook then **boosts routing**: proven-efficient models get
|
|
109
|
+
preferred under the `value` objective (`reason: proven 54% fewer tokens on edit`).
|
|
110
|
+
`poly sync` uploads *only* the playbook by default — your goals and raw logs never
|
|
111
|
+
leave the machine unless you pass `--raw`.
|
|
112
|
+
|
|
113
|
+
### Local LLMs (Ollama / LM Studio) — $0 routing
|
|
114
|
+
|
|
115
|
+
```bash
|
|
116
|
+
ollama serve # or LM Studio's local server
|
|
117
|
+
poly config local on # default base: http://localhost:11434/v1
|
|
118
|
+
poly config local on --base http://localhost:1234/v1 # LM Studio
|
|
119
|
+
poly models -s local/ # local models join the catalog at $0
|
|
120
|
+
poly run "..." # cheapest objective → local wins what it can
|
|
121
|
+
```
|
|
122
|
+
|
|
123
|
+
Local models appear as `local/<name>`, cost $0, and need **no API key** — with
|
|
124
|
+
`local on` and no OpenRouter key, Polymath runs fully offline on your machine.
|
|
125
|
+
Tokens are still tracked, so the playbook learns when your local model is the
|
|
126
|
+
most efficient approach.
|
|
98
127
|
|
|
99
128
|
### Routing objectives
|
|
100
129
|
|