A language-model plugin for tailoring, local first, off by default, every suggestion confirmed by the person #1105

New issue

Open

opened 2026-10-30 21:25:20 +00:00 by tiagoagueda · 0 comments

tiagoagueda commented

2026-10-30 21:25:20 +00:00

Owner

docs/PLAN.md keeps one item of the old "after version 1" list unbuilt: assistance from a language model for tailoring, as a plugin, off by default, never required. The wiki's Roadmap lists it as still wanted. This issue gives it a shape that keeps to the rules.

What exists today

No model is inside Postulo, by decision (PLAN.md section 2). postulo-mcp lets an outside agent read and, if allowed, write, and wraps third-party text in "untrusted" markers.
#241 item 5 proposes a local-model helper (Ollama or llama.cpp) for classifying mail and reading postings.
postulo-imap classifies mail with phrase lists in eight languages and skips the rest.

What would change

A plugin with a connection to a model the person names: a local server by default, any OpenAI-compatible address if they choose.
What it offers, each a suggestion the person accepts or not:
- a draft paragraph for a letter, from the listing and the career record;
- which career entries to put in a CV variant for a listing;
- a reading of a mail in a language postulo-imap has no phrase list for.
It never runs on its own. Every call is a button press, and the page says what text goes where.

Decisions

Local only, or any endpoint. Recommended: local by default; any endpoint allowed, with the connection naming where the text goes.
One plugin or several. Recommended: one, with features switched on one by one.
Untrusted text. Recommended: listings and mail wrapped as postulo-mcp wraps them; the model's answer is a suggestion only, never written by itself.

Checked against main at daee30b15 (2026-10-30). From the status review and feature suggestions of 2026-10-30.

`docs/PLAN.md` keeps one item of the old "after version 1" list unbuilt: assistance from a language model for tailoring, as a plugin, off by default, never required. The wiki's *Roadmap* lists it as still wanted. This issue gives it a shape that keeps to the rules. ## What exists today - No model is inside Postulo, by decision (PLAN.md section 2). postulo-mcp lets an outside agent read and, if allowed, write, and wraps third-party text in "untrusted" markers. - #241 item 5 proposes a local-model helper (Ollama or llama.cpp) for classifying mail and reading postings. - postulo-imap classifies mail with phrase lists in eight languages and skips the rest. ## What would change - A plugin with a connection to a model the person names: a local server by default, any OpenAI-compatible address if they choose. - What it offers, each a suggestion the person accepts or not: - a draft paragraph for a letter, from the listing and the career record; - which career entries to put in a CV variant for a listing; - a reading of a mail in a language postulo-imap has no phrase list for. - It never runs on its own. Every call is a button press, and the page says what text goes where. ## Decisions - **Local only, or any endpoint.** Recommended: local by default; any endpoint allowed, with the connection naming where the text goes. - **One plugin or several.** Recommended: one, with features switched on one by one. - **Untrusted text.** Recommended: listings and mail wrapped as postulo-mcp wraps them; the model's answer is a suggestion only, never written by itself. Checked against `main` at `daee30b15` (2026-10-30). From the status review and feature suggestions of 2026-10-30.

tiagoagueda added this to the 1.0.0 milestone

2026-10-30 21:25:20 +00:00

tiagoagueda added the

enhancement

label

2026-10-30 21:25:20 +00:00

tiagoagueda referenced this issue

2026-10-30 21:25:22 +00:00

Plugins worth writing next: ATS job-board sources, a WebDAV store, a LinkedIn export importer, modern mail sign-in, real templates #241