Skip to content

Feat/prompt cache gpt6.2 - #5

Merged
noxgle merged 2 commits into
mainfrom
feat/prompt-cache-gpt6.2
Nov 18, 2026
Merged

noxgle merged 2 commits into
mainfrom
feat/prompt-cache-gpt6.2

Conversation

@noxgle

@noxgle noxgle commented Nov 18, 2026

Copy link
Copy Markdown
Owner

No description provided.

Enable Chat Completions tool calling for both providers and accept
DeepSeek DSML text tool calls as a fallback. Also read OpenRouter's
non-streaming reasoning field and surface mid-stream provider errors
instead of returning an empty successful answer.

Bump version to 0.1.146.
- Register gpt-6.2-sol/gpt-6.2-astra with a known ~1.05M context window
- Move the system prompt into the first developer message (GPT-6.2 never
  caches top-level instructions); keep instructions as a short constant
- Add providerProfiles.<p>.promptCache settings (enabled/ttl/stableBreakpoint/
  toolResultBreakpoints/key/prewarm) with validation; reject legacy 24h TTL
- Send prompt_cache_options + explicit prompt_cache_breakpoint when enabled;
  account key derives a stable non-secret hash, never per-session
- Typed PromptCacheUnsupportedError on 400/422 with one bounded retry without
  cache fields and a session fallback flag
- Map xhigh->xhigh reasoning effort for 6.2; omit temperature/top_p for the family
- Sort native tool definitions for a byte-stable cache prefix; emit prefixHash
  on provider request events; warn when toolOutputPruning conflicts with cache
- Split cache-write telemetry by TTL (cacheWrite/cacheWrite30m/cacheWrite12h),
  add nonCachedInput and a separate 12h write price rate
- One-time prompt_cache_options.prewarm at first prompt (best-effort)
- Tests: settings validation, payload flags, fallback, prefix stability,
  compaction prefix reuse, telemetry split, model catalog; docs + CHANGELOG
  (no live benchmark performed)
@noxgle
noxgle merged commit 991f501 into main Nov 18, 2026
5 of 9 checks passed
@noxgle
noxgle deleted the feat/prompt-cache-gpt6.2 branch November 18, 2026 18:27
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant