A local endpoint that publishes no limits — LM Studio's /models says nothing —
left every model on the 8192/4096 defaults with no way to say "this one has a
131k window". Each discovered model now gets three optional pins in the
endpoint editor: context window, max output tokens, thinking (auto/on/off).
A pin always wins: user pin → what the endpoint published → compiled default.
Blank means no opinion, so discovery keeps ruling wherever you don't pin. Pins
live in customProvider.overrides keyed by model id, stored apart from the
discovery snapshot: a reconnect refreshes the models and prunes the pins of
models that vanished while keeping the rest.
Pins are written by PATCH /api/custom-provider, which never probes — the whole
point is configuring a server that isn't answering /models correctly. Connect
still probes and still fails loudly when the endpoint is down. A pin under 2048
tokens is rejected on the write side and clamped up on the read path, matching
the compaction-floor discipline.
Custom models were pinned reasoning:false, so every one of them offered a
single thinking level ("off") while built-in reasoning models offered the
standard ladder. The endpoint's /models metadata now decides:
- detectReasoning() trusts explicit metadata only — a boolean
reasoning / supports_reasoning / thinking, or a reasoning token in
supported_parameters (vLLM, OpenRouter), capabilities, tags or
architecture.output_modalities. No name guessing: a false positive would
send reasoning_effort to a server that rejects it.
- CustomModel carries the detected flag (persisted, sanitizer keeps it),
toPiModel passes it through, and pi-ai does the rest — default
off/minimal/low/medium/high set, reasoning_effort on the wire.
Verified in the browser against a local OpenAI-compatible mock: the
reasoning-capable model exposes off/minimal/low/medium/high, the plain one
stays at off, and a prompt with medium actually arrives as
reasoning_effort=medium.
The prior change made the vision slot show every model (image-capable first),
which relaxed the filter for built-in providers too — built-in text-only models
became selectable for Vision. Centralize the vision rule in ModelMenuList:
built-ins must report image input; custom-endpoint models stay selectable (their
capability may be unreported). Image-capable still float first.
Replace arbitrary event-loop budgets in two release-gate tests with fresh per-attempt sign-in notifications and the actual consent control reply, bounded by a clear watchdog. Retain all final-state assertions and explicitly verify successful connected/error-free completion.
Controlled token/save gates demonstrate both old 100-turn waits return before completion; the corrected waits remain pending and retain the final persistence/UI assertions after release. Add small watchdog result/error/timeout tests. No production, workflow or dependency changes.
Isolated checks: focused 87/87, full suite once 2517/2517, full ESLint and typecheck. The full run preceded three runtime-erased non-null array-index annotations; focused tests/typecheck/full ESLint passed again on final source.
Preserve original contributor commit 3bb28e5f29. Resolve only the settings-dialog conflict by retaining current General/update behavior: the donor General-only language placement is explicitly superseded by the human-approved row BELOW the full-width Deepgram key. The assigned implementation worker will implement that placement and approved shared dropdown finishing. Foundation only, not a checked candidate.
Adapt the bounded probe-completion foundation from PR #130 by aksalatdev (source commit 535202bebe038fa1e0648f5df809d5ccfc26d071), not its mixed stream-salvage implementation.
Require a non-failed call followed by a matching probe_done result with neither isError nor explicit details.ok:false. Preserve older clean acknowledgments, saved attempts, and all existing handoff/retry/provider policies. Share the native completion seam with phase detection and exclude explicit refusals from UI phases.
Verify retained failed-probe Retry recovery with real agent/tool composition and synthetic engines: one fresh conclusion, one handoff, then teacher selection, without stored-tool replay. Focused tests 58/58; full tests 2473/2473; typecheck and full ESLint pass in isolated storage with scrubbed provider credentials.
Co-authored-by: aksalatdev <73862605+aksalatdev@users.noreply.github.com>
Avoid hardcoding Promise<void> so the test also typechecks with PR154's Promise<boolean> admission result. Type-only correction; no product behavior changes.
A failed-response Retry submits no composer text. Retire the earlier accepted text send's recovery fallback synchronously so a Retry model/key refusal cannot restore it over the current unsent draft. Never clear a later ordinary send's fallback at async settlement.
Add key/model503, superseding-draft and ordinary-refusal coverage. Verified 183 focused tests, typecheck, full ESLint, production build, and a red/green live production regression plus ordinary-draft recovery; no external provider calls.
Decode only Retry's HTTP 400 error field for the existing banner. Preserve raw model/key refusal envelopes and other submission-error handling.
Verified on the accepted-main integration: 2424 full-suite tests, typecheck, full ESLint, production build and isolated synthetic production-browser Retry/Submit smoke. No real provider calls.
Keep failed attempts in the saved transcript and omit only trailing failures from Retry's temporary agent history. Use the matching temporary prefix length when persisting genuinely new output.
Preserve the error notice across plan/compaction markers, refuse Retry after later compaction, and avoid synthesizing results for tool calls inside failed generations while retaining Pi's native filtering.
Verified: 173 focused Retry/session/compaction-projection/practice-record tests, typecheck, changed-file lint, and diff --check. No app launch or real provider calls.
new URL(import.meta.url).pathname keeps percent-encoding, so the test
failed with ENOENT when the checkout path contains a space. Use
import.meta.dirname like the other tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A long first message in ?prompt= exceeds Node's header limit. The chat GET returns 431 and Electron commits an empty window. Stash the text in sessionStorage and open /chat/<id>.
Daniel Buckner