8ddcb5ca13
* fix(kimi-code): persist picked thinking effort up to the model's default effort The persistence gate kept the model's top declared effort session-only unconditionally, so users whose delivered default_effort is the top tier (e.g. max) could never save an explicit pick of it. Compare the pick against the model's default_effort instead, using support_efforts as the strength ordering: picks above the default stay session-only, picks at or below it persist. Models without a declared default keep the historical top-tier rule. The same change lands in the VS Code extension's mirrored logic. * docs(kimi-code): document the effective-default ceiling for effort persistence Clarify in both apps' comments, the changeset, and the config docs that the persistence ceiling is the model's effective default effort, whether declared via the catalog / overrides or synthesized by the protocol profile inference (Claude models resolve to high, so an xhigh pick is session-only there). Pin the inference path with tests in both apps. * fix(vscode): resolve the save-config model with its provider type Mirror the TUI's effectiveModelForHost: without the provider type the Anthropic fallback profile (e.g. claude-latest) never matches, so the inferred default effort that gates persistence was missed and an above-default pick could persist where the TUI keeps it session-only. * Delete .changeset/persist-effort-up-to-model-default.md Signed-off-by: 7Sageer <sag77r@hotmail.com> * fix(vscode): seed the persisted effort with the effective-default ceiling * fix(vscode): project webview models with the provider type * fix(kimi-code): apply a session-only effort pick to the runtime in the /provider flow * fix(vscode): update the persisted-effort seed on model-switch saves * docs(kimi-code): drop the default_effort persistence-ceiling note --------- Signed-off-by: 7Sageer <sag77r@hotmail.com> |
||
|---|---|---|
| .. | ||
| .vscode | ||
| docs | ||
| resources | ||
| scripts | ||
| shared | ||
| src | ||
| test | ||
| webview-ui | ||
| .vscode-test.mjs | ||
| .vscodeignore | ||
| CHANGELOG.md | ||
| LICENSE | ||
| package.json | ||
| README.md | ||
| tsconfig.json | ||
| tsdown.config.ts | ||
| vitest.config.ts | ||
Kimi Code
AI coding assistant for VS Code, built for long-context workflows and complex coding tasks.
Features
- Works alongside you: Kimi autonomously explores your codebase, reads and writes code, and runs terminal commands with your permission
- Thinking controls: Toggle reasoning or choose a model-supported thinking effort
- Provider-aware models: Distinguish and select same-named models across configured providers
- Native editor integration: Review AI-proposed changes directly in VS Code's diff viewer
- MCP support: Extend capabilities with Model Context Protocol servers
- Slash commands: Quick actions like
/initto analyze your project and/compactto manage context
Install
Kimi Code requires VS Code 1.100.0 or later.
- Install from the VS Code Marketplace
- Open a folder in VS Code
- Click the Kimi icon in the Activity Bar
- Sign in with a kimi.com/code subscription, or use a provider already configured in the shared
config.toml
The extension runs the Kimi Code Node SDK in the VS Code Extension Host. When
the extension and the Kimi Code terminal app resolve to the same
KIMI_CODE_HOME, they share config.toml, MCP configuration, login state, and
sessions. The system-level KIMI_CODE_HOME environment variable is supported;
there is no separate VS Code setting for it. Do not run the same session from
both applications at the same time, because cross-process session locking is
not guaranteed.
After upgrading from version 0.5.x, the extension prompts before migrating any legacy data it finds. Migration copies or merges data into the current Kimi Code home and does not delete the legacy source. Legacy Kimi Code OAuth and MCP OAuth credentials are not copied, so those connections must be authorized again. See the changelog for the full compatibility notes.
Docs
Official doc for Kimi Code can be found at www.kimi.com/code/docs