Days elapsed since `modified_secs`. Floor-rounded; clamped to 0 for future mtimes (clock skew). Mirrors `memoryAgeDays` in `memoryAge.ts`.

Plain-text staleness caveat for memories > 1 day old. Returns an empty string for fresh memories (today / yesterday). Mirrors `memoryFreshnessText` i

Merge remote settings on top of local settings. Remote settings take precedence for any key they define (enterprise policy override). Local settings

Mistral AI — Reads `MISTRAL_API_KEY`. Uses OpenAI-compatible format with Mistral-specific quirks: - Tool call IDs must be alphanumeric only, truncated

Add a regional cross-inference prefix for models that support it.

Returns `true` for models that support the maximum effort tier. For Claude this is Opus-only; for the GPT-5 family the top tier maps to OpenAI's `xhi

Resolve the models.dev source URL, honoring env-var overrides.

Moonshot AI / Kimi. Reads `MOONSHOT_API_KEY`.

`b` — start of previous word.

`e` — end of current/next word.

`w` — start of next word.

Move cursor left.

Move cursor right.

Move the cursor to the same visual column on the row below. Returns `true` if the cursor actually moved (i.e. there was a row below).

Move the cursor to the same visual column on the row above. Returns `true` if the cursor actually moved (i.e. there was a row above).

Alt+B: Jump to previous word.

Alt+F: Jump to next word.

For multi-enum fields: move the sub-cursor down.

Nebius — Llama / Qwen hosting. Reads `NEBIUS_API_KEY`.

Cycle to next status in the UI flow: pending→in_progress→completed→pending. Running, Failed, and Deleted are treated specially (skip to Pending when c

Normalise a line for case-insensitive token matching. Collapses multiple spaces / tabs into a single space and trims.

Normalize a URL: strip query-string and trailing slash.

Novita — Llama / Stable Diffusion hosting. Reads `NOVITA_API_KEY`.

Nvidia NIM — enterprise AI inference. Reads `NVIDIA_API_KEY`.

Bind to an ephemeral localhost port for the OAuth redirect. Returns the allocated port number.

Record a successful compaction.

Open the rewind flow with the current message list converted to `SelectorMessage` entries.

Open the overlay with full state context.

OpenCode Go — flat-rate subscription endpoint hosted by opencode.ai. OpenAI-compatible chat completions surface; same key works for the Zen metered ti

Total number of selectable rows: options + custom-text row.

OVHcloud — EU-hosted AI. Reads `OVHCLOUD_API_KEY`.

Parse the events map `{ "PreToolUse": [ { matcher: ..., hooks: [...] } ] }`.

Parse a JSONL transcript with std::fs, returning the non-tombstoned entries.

Convert an OpenAI ChatCompletion response to Anthropic format fields. Returns (content_text, finish_reason, input_tokens, output_tokens)

Internal: return `(id, matched_string, byte_offset)` for each reference.

Extract content blocks and usage from a completed Gemini response body.

Perplexity — search-augmented LLM API. Reads `PERPLEXITY_API_KEY`.

Persist all durable tasks to `~/.claurst/scheduled_tasks.json`.

Generate a PKCE code verifier (43 URL-safe random chars per RFC 7636).

Remove the last character from the rename input buffer.

Go back to the previous page.

Return the project-level plugins directory: `<project>/.claurst/plugins`.

Push tagged lines from `render_transcript_assistant_message_tagged`. Lines with `Some(hash)` become clickable thinking headers.

Qwen / Alibaba DashScope. Reads `DASHSCOPE_API_KEY`. Uses a default temperature of 0.55 as recommended by Alibaba's docs.

------------------------------------------------------------------ Internal: produce 32 random bytes. We derive them from a UUID v4 (which already pul

Read text from the system clipboard. Returns `None` if the clipboard is empty, unavailable, or contains non-text data.

Cheap tail read to pull `last-prompt` and `custom-title` without a full parse — used when we don't need to walk all messages.