Context budget calculator
Pick a hardware tier (T0–T9) for preview and a target
chat.max_context_tokens (500–200,000). You choose the value in YAML; defaults ship at install.
Generation length uses chat.max_gen_tokens
(operator ceiling for Ollama
num_predict) — Engine desktop Apply co-writes both keys for your tier.
Impact warnings are advisory only.
config/contextmint.defaults.yaml on the API host, or overlay
~/.contextmint/server.defaults.yaml.
Live values: GET /api/v1/config/context-budget?max_context_tokens=….
Tier in this page models runtime ChatCaps from detector.py — not a config validator.
Inputs
Valid range 500–200,000 tokens (server schema max). Presets are shortcuts; type any value in range.
Scaling matrix
Derived chat.* fields for your selected
max_context_tokens when
context_budget.auto_scale is true (ratios from shipped defaults; baseline
10000).
| Server variable | Calculated value | Formula |
|---|
Tier envelope (ChatCaps)
Startup probe maps RAM/VRAM → tier → these ceilings. Values from detector.py.
Effective RAG budget (selected tier)
Runtime order: config cap → hardware clamp → intent floor (raised up to config cap). Health-policy overrides are not modeled here.
Tier comparison matrix
Effective RAG chunk budget per tier for the same max_context_tokens.
Ratio profile (edit in YAML)
| Ratio key | Fraction of max_context_tokens |
|---|
Example overlay
Live apply (Engine)
Connected to . Writes
~/.contextmint/server.defaults.yaml via loopback API — server restart required.