v0.9.5

[0.9.5]

Fixed

• Improved CLI error messages so authentication, validation, and network failures explain what was not completed and how to recover.

v0.9.4

[0.9.4]

Fixed

• kagi assistant prompt mode now reads a missing QUERY from non-empty stdin, including streamed prompts such as echo "hello" | kagi assistant --stream.

v0.9.3

[0.9.3]

Fixed

• Hardened demo scripts so they no longer use a predictable shared /tmp PATH shim before running authenticated kagi commands.
• Made Unix and PowerShell installers replace existing binaries through staged same-directory writes instead of overwriting the installed binary directly.
• Tightened release, publish, security, coverage, CI, and Makefile checks for more reproducible and complete release validation.
• Isolated integration test home and XDG directories so local user config cannot leak into test runs.
• Reduced default CLI failure output to a single user-facing diagnostic.

v0.9.2

[0.9.2]

Added

• Added kagi skills and kagi agent so agents can load embedded, version-matched CLI usage guidance directly from the installed binary.
• Added Mintlify command documentation for the new embedded skill workflow.

v0.9.1

[0.9.1]

Changed

• Search auth routing now preserves session-preferred fallback behavior when API credentials are unavailable or rejected.
• Lens create, update, enable, and disable flows now validate lens names more consistently and behave more reliably against live Kagi accounts.
• Assistant examples and live-test coverage now use the current model catalog.
• Release and package metadata were updated for the current GitHub Actions workflow shapes.

v0.9.0

[0.9.0]

Added

• Added kagi completion generate and kagi completion install to generate or install shell completions for Bash, Zsh, Fish, and PowerShell.
• Added configurable Assistant streaming output. kagi assistant --stream now writes incremental text deltas by default, and --stream-output json keeps structured newline-delimited JSON events.

Changed

• Base search in API-first mode now falls back to the session-token path when the Search API rejects the API key, including rate-limit and quota-style failures.
• Updated README and Mintlify docs for auth routing, completion installation, and Assistant streaming behavior.

v0.8.1

[0.8.1]

Added

• Added V1 Search API support for --region, --from-date, --to-date, and --limit when search is routed through KAGI_API_KEY.
• Added a release workflow step that triggers the Mintlify docs deployment when MINTLIFY_DEPLOY_COOKIE is configured.

Changed

• Clarified current /api/v1 API key behavior and legacy /api/v0 API token behavior across README and Mintlify docs.

v0.8.0

[0.8.0]

Added

• Added kagi assistant models for stable JSON output of available Assistant base-model slugs.
• Added kagi assistant --stream to emit NDJSON updates with md_delta while Assistant responses are generated.
• Added kagi assistant --once --model <MODEL> to create a temporary custom assistant for one prompt and delete it afterward.

Changed

• kagi extract and MCP kagi_extract now require KAGI_API_KEY directly instead of trying to derive an API key from session auth.

v0.7.0

[0.7.0]

Added

• Added KAGI_API_KEY, [auth].api_key, and kagi auth set --api-key for current /api/v1 Search and Extract API credentials.

Changed

• Breaking: split current API keys from legacy API tokens. KAGI_API_TOKEN and [auth].api_token now represent legacy /api/v0 credentials only, while base Search API mode requires KAGI_API_KEY or [auth].api_key.

v0.6.1

[0.6.1]

Changed

• Updated serde_json to 1.0.150.