Memory

Long-term Memory

Long-term memory is an optional feature that lets TypeWhisper remember durable facts from your dictation so future prompt actions can recall them as context.

How it works

When an IMemoryStoragePlugin (for example FileMemory or OpenAiVectorMemory) is enabled alongside a configured LLM provider, eligible transcriptions are sent to the LLM to extract durable facts. Those facts are stored by the memory storage plugin, and future Prompts & AI actions can recall them as context.

Memory storage plugin	Storage
FileMemory	Local JSON.
OpenAiVectorMemory	Embedding-backed recall.

See Other plugins for the memory storage plugins.

Eligible transcriptions

Not every dictation triggers extraction. A transcription is eligible only when:

• It is at least 30 characters long — shorter dictations are skipped.
• At least 30 seconds have passed since the last extraction this session — a per-session cooldown so extraction does not fire on every short dictation.

What gets extracted

Eligible transcriptions are sent to the LLM, which extracts lasting personal facts:

• Names, job titles, preferences, locations, relationships, projects, tools used, responsibilities, and recurring topics.

Temporary information such as meeting times and deadlines is explicitly excluded. If no lasting facts are found, nothing is stored.

Automatic recall

Recall is automatic. When memory is enabled, every prompt action looks up the stored facts most relevant to the input and injects up to the top 10 matches into that action's system prompt as remembered context before the LLM runs.

Enabling memory

The Advanced settings page has a memory enable toggle. The toggle is gated on having both a memory storage plugin and an available LLM provider — without both, the toggle is unavailable.

Related pages

• Other plugins — the FileMemory and OpenAiVectorMemory storage plugins.
• Advanced settings — the gated memory enable toggle.
• Prompts & AI actions — prompt actions that recall stored facts as context.

---

Changelog

Date	Change
2026-06-17	Initial version.
2026-06-17	Documented eligibility (30-char minimum, 30-second cooldown), what is extracted/excluded, and automatic top-10 recall into prompt actions.