Kilo isn't always BYOK; what matters here is that Kilo doesn't charge a
per-request fee on top of the upstream inference cost. So for the Kilo
provider, the gateway/marketplace `cost` would understate the user's true
spend — always prefer the upstream/market cost when present.
* feat: support Kilo-hosted indexing embeddings
* fix: format indexing settings
* fix: serialize local model persistence
* fix: annotate local model persistence
* fix: stabilize MCP OAuth browser test
* fix: annotate MCP OAuth test wait
* fix: keep KiloProvider under line cap
* fix: align Kilo indexing auth defaults
* fix(vscode): add missing indexing translations
* feat(indexing): fetch Kilo embedding model catalog
* fix(indexing): align Kilo embedding metadata fallback
* fix(gateway): remove unused indexing dependency
* fix(indexing): keep fallback catalog webview safe
* fix(agent-manager): include embedding model provider
* fix(vscode): provide embedding models in stories
* refactor(indexing): rely on Cloud embedding catalog
* fix(indexing): hydrate Kilo model metadata from Cloud
* fix(indexing): clear model when switching provider
Refactor providerCost to read `usage.raw.cost_details.upstream_inference_cost`,
which both `@ai-sdk/anthropic` (`convertAnthropicUsage`) and `@ai-sdk/openai`
(`convertOpenAIResponsesUsage`) populate with the verbatim OpenRouter usage
object. This unifies the Anthropic Messages and OpenAI Responses cost paths
through a single branch and removes the redundant `metadata.anthropic.usage`
branch (the SDK exposes the same data under both keys for Anthropic, and only
under `raw` for Responses).
They're upstream opencode comments, not kilo changes, so they
shouldn't live inside kilocode_change markers — and outside the
markers they add noise to this PR's diff. Drop them.
Add missing kilocode_change marker to the editorContext param in
the environment Effect.fn implementation; the interface signature
already had a marker but the implementation didn't.
For the Anthropic Messages API (via OpenRouter or Vercel AI Gateway), the
top-level `cost` field is the gateway fee that Kilo paid (typically 0 for
BYOK), not the user's true spend. Only use `upstream_inference_cost` /
`marketCost`; fall back to the token-based calculation when those are
absent.
Allow Azure provider connections to capture resource names or endpoint URLs so GPT-5 deployments can use the native Azure path instead of generic OpenAI-compatible setup.
Cost reporting already worked when Kilo used OpenRouter chat completions
internally. Extend providerCost to also read cost from the Anthropic
Messages API stream metadata, covering both OpenRouter (Anthropic-style
`usage.cost` / `cost_details.upstream_inference_cost`) and Vercel AI
Gateway (`gateway.cost` / `gateway.marketCost`).
Revert the file-wide marker shape fixes from fix-kilocode-markers.ts
since they touch unrelated code. Scope this PR to the two review
findings in the gpt-5 block:
- Replace nativeOpenAI refactor with upstream's if (===||===) form
plus inline // kilocode_change markers on the added npm entries;
restores the upstream reasoningSummary comment.
- Restore the upstream textVerbosity comments, keep the new provider
gate wrapped in kilocode_change start/end.
- Reverts the nativeOpenAI refactor; restores upstream's reasoningSummary
guard structure + comment so added npm entries stay inside a single
marker block.
- Includes the opencode textVerbosity comment that was dropped.
- Also fixes pre-existing marker shape issues elsewhere in the same file
flagged by script/upstream/fix-kilocode-markers.ts.
Minimizes the diff from upstream opencode by restoring upstream's
original reasoningSummary guard structure and comment, and scoping
kilocode_change markers to only the Kilo-specific additions:
two extra provider npm names and the new textVerbosity provider gate.
Keep upstream's `question` tool binding and `questionEnabled` boolean
names, restore upstream's `tool.plan` position at the end of the builtin
array, and group Kilo-only additions in a single marker block. Removes
outer markers that were wrapping unchanged upstream lines as Kilo-owned
scope.
Replaces the plugin approach with a narrow conditional at the single call
site in session/llm.ts where maxOutputTokens is passed to chat.params.
No new file, no new test file, one kilocode_change block.
gpt-5 models rejected via openai-compatible proxies (e.g. LiteLLM) because
the SDK emits max_tokens while OpenAI requires max_completion_tokens and
the compatible SDK cannot rename the field. Clear the cap so the upstream
default output budget applies.
* fix(vscode): restore inline tool diffs
* fix(vscode): render tool patches in kilo ui
* style: format long regex assignments and add change marker comment
Reformat multi-line regex match assignments in kilo-ui-contract test
to satisfy line length limits, and annotate the `contents(diff)` call
in session-diff with a kilocode_change tracking comment.
* fix(kilo-vscode): append trailing newlines to expected diff content assertions
Update test expectations in diff-session-source to include trailing
newlines in before/after content, matching actual file content behavior.
* fix(vscode): guard empty-patch diffs in session turn accordion
Match diff-session-source.ts:99 behavior by short-circuiting contents()
when the patch is empty (binary or summarized files), so the accordion
content stays empty instead of rendering a confusing whitespace-only
diff.
---------
Co-authored-by: Imanol Maiztegui <imanol.mzd@gmail.com>
Upstream 2283979199 (Preapprove agent tmp directory access) extended the
agent.ts whitelistedDirs with Global.Path.tmp/*, and the new v1.14.33 test
asserts tmp/agent-work -> allow on the explore agent. Kilo's patchAgents
replaces the whole explore permission and had only Truncate.GLOB in its
external_directory block, so both /some/other/path (expected ask) and
tmp/agent-work (expected allow) resolved to deny via the outer '*':'deny'
catch-all (findLast picks it over any defaults external_directory rule).
Commit d2e21c5006 tried to fix this by dropping the '*':'ask' on the
assumption that defaults already provided it — that rationale was wrong,
since defaults' rules come before the patch's catch-all in the merged
ruleset. Mirror upstream's inline shape instead: thread whitelistedDirs
through to patchAgents and rebuild the explore external_directory as
{ '*':'ask', ...whitelistedDirs -> 'allow' }, matching what upstream's
explore does natively.
The previous `watchTelemetryState` fix updated the webview UI in real
time but left the CLI subprocess's PostHog client stuck on its
spawn-time `KILO_TELEMETRY_LEVEL` value. A user who started VS Code
with telemetry off and toggled it on at runtime saw the thumbs UI
appear (good) but every webview event was silently dropped at the
CLI's `Client.capture()` gate (bad).
Add a runtime sync channel:
- New `POST /telemetry/setEnabled` Hono route on the CLI server that
calls `Telemetry.setEnabled(enabled)` to flip the `posthog-node`
client's opt state.
- New `TelemetryProxy.setEnabled(enabled)` method that POSTs to it,
using the same fire-and-forget pattern as `capture`.
- Extension calls `telemetry.setEnabled(vscode.env.isTelemetryEnabled)`
immediately after `telemetry.configure(...)` on every `connected`
state change, so a freshly-spawned CLI gets corrected even when its
spawn-time env var is stale.
- Extension subscribes to `vscode.env.onDidChangeTelemetryEnabled` to
forward runtime consent changes to the CLI as they happen.
All three changes live in Kilo-owned files (the route is already a
`kilocode_change - new file`, and the extension is Kilo-only). Zero
upstream OpenCode merge surface.
Closes#9872 fully (the previous `d67c5e307c` covered only the webview
UI).
Addresses review feedback: Kilo-specific additions belong in Kilo-owned
packages so the diff against upstream OpenCode stays minimal.
- Icons: move thumbs-up/-down/-up-filled/-down-filled paths from
packages/ui/src/components/icon.tsx into packages/kilo-ui's icon
registry. Extend the kilo-ui Icon registry to carry per-icon viewBox
(Heroicons thumbs are 20x20; existing Kilo icons are 16x16). Replace
packages/kilo-ui/src/components/icon-button.tsx (previously a re-export
of @opencode-ai/ui/icon-button) with a local implementation that
uses the kilo-ui Icon so the new names resolve.
- i18n: move the three feedback strings (helpful, notHelpful,
clearRating) from each of 19 packages/ui/src/i18n/<locale>.ts files
into the corresponding packages/kilo-i18n/src/<locale>.ts. The
webview's language.tsx already merges kilo-i18n on top of upstream,
so the runtime keys remain available.
- TUI: extract submitFeedback from
packages/opencode/src/cli/cmd/tui/routes/session/index.tsx into a
new Kilo-owned helper at
packages/opencode/src/kilocode/cli/cmd/tui/feedback.ts. The session
route now imports and invokes it, passing { toast, session, messages }.
Also tightens the Kilo Gateway gate from startsWith("kilo") to
=== "kilo" to match the equivalent fix in the webview.
- Replace <leader>+ with <leader>= in messages_feedback_up: the TUI
parser normalizes <leader> to 'leader+' and splits on '+', producing
an empty key name for '<leader>+' so the binding never fires. The
= key sits on the same physical key as + on most layouts, pairs
visually with <leader>-, and parses to a real key name.
- Wrap StoryProviders with FeedbackProvider so stories that render
VscodeSessionTurn (e.g. Diff Summary Collapsed) don't throw
'useFeedback must be used within a FeedbackProvider' under
Storybook visual regression.
- Sync changeset and docs page to mention the new keybind.
Adds thumbs up/down buttons next to the copy button on every assistant
message in the VS Code sidebar, and <leader>+/<leader>- keybinds in the
TUI. UI state is in-memory only — ratings reset on reload / session
switch. Persistence can be added later without changing the telemetry
contract.
Events are sent to PostHog via the existing telemetry pipeline. For
Kilo Gateway turns the payload includes session and message IDs so
feedback can be correlated against gateway logs; for direct providers
those IDs are omitted since we cannot correlate them to upstream data.
- InstanceStore: run Instance.provide init inside the ALS context so KilocodeBootstrap (and any forkDetach work it spawns like KiloIndexing.init) can read Instance.directory. Upstream refactor moved init out of the ALS scope; kilo-main's pre-merge Instance.provide wrapped it. Without this, KiloIndexing.init silently fails with "No context found for instance".
- kilocode/agent: thread worktree through planGuard/patchAgents instead of reading Instance.worktree at agent state construction time. Agent state is built inside Effect (no ALS) via InstanceState.make — reading the ALS-backed Instance.worktree there crashed under the new architecture.
- kilocode/agent: drop the "*": "ask" from explore external_directory — defaults already provides it, and redefining here overwrites the tmp/skill allowlist via findLast().
- test/server/httpapi-instance.test.ts: revert the ported Hono-bridge tests. Upstream put the same tests in httpapi-instance.legacy.test.ts (renamed file); the port duplicated them.
- test/server/httpapi-instance.legacy.test.ts: mark the catalog test test.skip with Kilo's original rationale (/agent 500s via the bridge; the bridge is not enabled in any production client).
- test/server/httpapi-ui.test.ts: delete. Tests upstream's proxy-to-app.opencode.ai fallback that Kilo intentionally removed (src/server/routes/ui.ts kilocode_change).
- test/server/httpapi-raw-route-auth.test.ts: basic("opencode", ...) → basic("kilo", ...) to match the Kilo username default.
- test/provider/models.test.ts: skip describe block. Upstream tests assert raw-fixture passthrough but Kilo's ModelsDev.get() filters/injects providers based on Config.get(), which needs an Instance context the test doesn't provide.
Replace 'as SessionPrompt.PromptInput' with 'as unknown as SessionPrompt.PromptInput' to match the pre-existing pattern in the legacy Hono handler (instance/session.ts:901,936). The single cast was rejected by TypeScript because the schema-derived ctx.payload has readonly arrays while Kilo's PromptInput override declares parts as mutable PartInputUnion[]. The double cast is the established Kilo workaround for this readonly→mutable mismatch and keeps the Kilo diff vs upstream minimal.
Mechanical follow-up to the v1.14.33 merge:
- src/server/routes/ui.ts: remove dead proxy fallback (proxy/createHash/csp were intentionally commented out by Kilo; the Effect/Promise paths still referenced them)
- src/kilocode/{plan-followup,session/prompt}.ts: pass Instance.current to Session.plan(input, instance)
- src/server/routes/instance/httpapi/handlers/session.ts: cast PromptPayload spread to PromptInput (readonly→mutable schema/runtime mismatch)
- src/plugin/index.ts: cast external @opencode-ai/plugin auth plugins through unknown to bridge to local @kilocode/plugin types
- script/build.ts: drop duplicate sourcemap key (Kilo's release-aware version wins)
- test/kilocode/indexing-{startup,worktree}.test.ts: switch from Instance.disposeAll/InstanceBootstrap-as-effect to disposeAllInstances/getBootstrapRunEffect
- test/kilocode/model-cache-org.test.ts: convert Instance.provide init from async fn to Effect
- test/kilocode/kilo-loader-auth.test.ts: drop ModelsDev.Data.reset() (no longer exists)
- test/kilocode/plan-exit-detection.test.ts, plan-followup.test.ts: pass Instance.current to Session.plan
- test/kilocode/{plan-followup,session-list}.test.ts: replace Session.list() with AppRuntime.runPromise(Session.Service.use((svc) => svc.list()))
Marian Alexandru Alecu