unread 0.1.0__py3-none-any.whl

presets/README.md

@@ -0,0 +1,96 @@
+# Presets
+
+Промпты для `unread analyze --preset <name>`. Каждый `.md` — один пресет.
+
+Файлы, начинающиеся с `_` (например `_reduce.md`), пресетами не считаются —
+это вспомогательные промпты для map-reduce.
+
+## Доступные пресеты
+
+| Пресет | Что выдаёт | Когда пригодится |
+|---|---|---|
+| `summary` | Короткое «что нового и неочевидного»: главное, идеи/решения, что посмотреть | Быстро понять, стоит ли читать чат глубже — дефолтный вариант |
+| `digest` | Короткий пронумерованный список тем, 1–2 строки на каждую | Быстрый сканирующий дайджест |
+| `action_items` | Markdown-таблица задач: кто/что/срок/статус/ссылка | Найти, что нужно сделать после чата |
+| `decisions` | Markdown-таблица решений: решение/кто/когда/обоснование/ссылка | «Что мы тут договорились?» |
+| `highlights` | 5–15 самых ценных сообщений, отсортированных по важности | «Не читать всё — покажи жемчуг» |
+| `questions` | Таблица открытых вопросов (без ответа / частично / без консенсуса) | Вернуться и ответить/обсудить |
+| `quotes` | Дословные памятные цитаты с автором и ссылкой | «Хорошо сказано» — сохранить |
+| `links` | Внешние URL из чата, сгруппированные по темам | Собрать ссылки на инструменты/статьи |
+
+Все пресеты выдают ссылки на конкретные сообщения вида
+`[#12345](https://t.me/username/12345)` (или `https://t.me/c/<id>/12345`
+для приватных чатов, с топиком — `.../thread/12345`), если Telegram
+разрешает построить ссылку. Для чатов без username и с положительным
+chat_id (личные переписки) ссылки не генерируются — пресет подставляет
+просто `#12345`.
+
+## Формат файла
+
+```markdown
+---
+name: summary                   # должно совпадать с именем файла
+prompt_version: v2              # бамп инвалидирует analysis_cache
+needs_reduce: true              # нужна ли reduce-стадия для map-reduce
+filter_model: gpt-5.4-nano      # модель для map-фазы (чанков)
+final_model: gpt-5.4            # модель для финального ответа / single-chunk
+output_budget_tokens: 1500      # max токенов на финальный ответ (reduce)
+map_output_tokens: 1500         # max токенов на чанк в map-фазе (по умолчанию 1500)
+---
+Текст system-промпта. Задаёт роль модели.
+
+---USER---
+
+Текст user-промпта (инструкция). Должен содержать плейсхолдеры:
+{period}, {title}, {msg_count}, {messages}.
+
+Период: {period}
+Чат: {title}
+Сообщений: {msg_count}
+---
+{messages}
+```
+
+## Форум-чаты
+
+Если анализируемый чат — форум, преамбула получает ссылку со вставкой
+thread_id: `https://t.me/{chat}/{thread_id}/{msg_id}`. Пресет про это
+знать не обязан: шаблон из преамбулы подставляется сам.
+
+Для форума есть три режима (на уровне CLI):
+
+- `--thread N` — один топик. Пресет видит только сообщения этого треда
+  и цитирует их через стандартный `#msg_id`.
+- `--all-flat` — весь форум «плоско». Пресет видит пересыпанные
+  сообщения всех топиков; ссылки идут без `/thread/`, но всё равно
+  ведут к конкретному сообщению.
+- `--all-per-topic` — по пресету на каждый топик. Отчёты складываются
+  в `reports/{chat-slug}/{topic-slug}-{preset}-{stamp}.md`.
+
+## Что видит модель в `{messages}`
+
+Формат сообщения в промпте:
+
+```
+[HH:MM #12345] sender_name [fwd: src] [voice 0:23]: ↩reply_sender text
+```
+
+- `#12345` — `msg_id`. Используй его в ссылках.
+- В преамбуле выше есть строка `Ссылка на сообщение: https://t.me/...{msg_id}`
+  — шаблон, куда подставить msg_id. Если её нет — значит чат не поддерживает
+  t.me-ссылки (личные переписки без username), пиши просто `#12345`.
+
+## Добавить свой пресет
+
+1. Скопируй любой файл (например `summary.md` → `my_preset.md`).
+2. Поменяй `name:` внутри на `my_preset`, отредактируй промпты.
+3. Запусти: `uv run unread analyze @chat --preset my_preset`.
+
+Для одноразового промпта без регистрации файла используй
+`--preset custom --prompt-file path/to/my.md` — тот же формат.
+
+## Кэш
+
+`prompt_version` в frontmatter — часть ключа `analysis_cache`. Пока он не
+меняется, повторные запуски того же пресета на тех же сообщениях идут из
+кэша (бесплатно). Бампай версию, когда меняешь промпты существенно.

presets/en/_base.md

@@ -0,0 +1,83 @@
+# Base instructions (shared across all presets)
+
+You analyze a stream of messages drawn from one source: a Telegram chat,
+a forum topic, the whole forum, or a video transcript. The specific task
+is described in the preset-specific section that follows below. These
+rules are the shared foundation.
+
+## Context from the preamble
+
+Before the messages comes a metadata block:
+
+- `=== Chat: <name> ===` (or `=== Video: <name> ===`) — name of the
+  chat / channel / forum / video.
+- `Period: …` — analysis time range (or "single video" for video mode).
+- `Forum: N topic(s) — …` — present only when the whole forum is being
+  analyzed (flat-mode). Lists every topic in the forum. If this line is
+  missing — you're working with a single chat, single topic, or video.
+- `Message link: <template>` — template for back-links to messages.
+  For videos, the template includes a timestamp parameter so citations
+  jump directly to the cited moment in the video.
+
+## Message format
+
+Each message is a single line:
+`[HH:MM #<msg_id>] <author>[tags]: <body>`
+
+For **video transcripts**: each line is a transcript segment, not a
+separate person speaking. The `<author>` is the channel name. The
+body usually starts with `[HH:MM:SS]` indicating the position in the
+video where this segment begins; `<msg_id>` is the same offset
+expressed in seconds — useful for citations that jump to the moment.
+
+### Tags in the header
+
+- `[voice MM:SS]` / `[videonote MM:SS]` / `[video MM:SS]` — voice or
+  video message with duration. Body is the audio transcript.
+- `[photo]` — photo without a description (vision enrichment is off).
+- `[fwd: <source>]` — a forwarded message.
+- `[reactions: 👍×3 ❤×1 …]` — reactions left by participants. A strong
+  signal of importance: messages with strong reactions are more likely
+  to deserve top placement and key bullets. But reactions are a hint,
+  not a substitute for substance. An empty reply with reactions doesn't
+  become valuable just because of them. Reactions don't apply to video
+  transcripts.
+
+### Inline body inserts
+
+- `[image: <description>]` — a vision-model description of the image
+  (present when image enrichment is on).
+- `[doc: <excerpt>]` — extracted document text (PDF / DOCX / code).
+
+After the body there may be lines of the form `  ↳ <url>: <summary>` —
+fetched-and-summarized external links. Treat them as context: they
+augment the message but are not "the author's own words".
+
+## Citation rules
+
+When citing a specific message, write `[#<msg_id>](<link>)`, where
+`<msg_id>` is the number from `#NNN` in the header, and `<link>` is
+formed by substituting `{msg_id}` into the template from the preamble.
+Without a template, just `#<msg_id>` with no hyperlink.
+
+For **video transcripts**, the `{msg_id}` is the second-offset of the
+cited segment, so a citation like `[#754](https://www.youtube.com/watch?v=ID&t=754s)`
+is a clickable jump to that moment. Prefer citing the moment when the
+relevant point is **made**, not where it's recapped.
+
+If the message stream contains `=== Chat: <name> ===` separators with
+their own `Message link: …` line (channel + comments mode), use the
+template **from the group the cited message belongs to**. msg_id is
+chat-local — identical numbers from different groups resolve to
+different links.
+
+## Writing rules
+
+- Write in English (or the source's language if it's clearly something else).
+- **Don't invent facts.** Rely only on the provided messages / segments;
+  don't reconstruct missing storyline.
+- Tight, no fluff. One bullet — one thought.
+- Cite messages / moments, don't describe their position in the conversation.
+- Skip noise: greetings, "ok", "thanks", lone emoji, paraphrases of
+  the same thought. For videos: skip filler sounds and verbal tics.
+- If little was said — write little. Don't pad for the sake of length.

presets/en/_reduce.md

@@ -0,0 +1,22 @@
+Below are several already-written mini-summaries of the SAME chat,
+produced from different conversation chunks. Merge them into **one**
+final summary in the requested format. Write in English.
+
+Merge rules:
+
+1. **Don't duplicate bullets.** If the same thought appears in multiple
+   chunks — combine into one bullet, gathering all relevant citations.
+2. **Preserve preset structure.** If the mini-summaries already have
+   sections (`##` / `###` / table headers), the result must have the
+   same structure. Don't flatten.
+3. **Preserve per-topic subsections.** If the mini-summaries split
+   content by forum topics (`### <topic name>`), **the final must do
+   the same**. Don't merge topics into one shared list.
+4. **Respect preset limits.** If the preset says "1–4 insights" or
+   "5–10 bullets" — apply the limit to the final (or to each
+   subsection if they exist), not to the merged sum. Drop middling
+   bullets rather than ship a 20-bullet wall.
+5. **TL;DR from mini-summaries appears in the final exactly once** —
+   pick the best variant or rewrite, don't concatenate them.
+6. Keep facts intact: numbers, names, thresholds, `[#NNN](...)` links —
+   verbatim, no creative reinterpretation.

presets/en/action_items.md

@@ -0,0 +1,52 @@
+---
+name: action_items
+prompt_version: v1
+description: Tasks from the chat — table of who/what/deadline/status
+needs_reduce: true
+filter_model: gpt-5.4-nano
+final_model: gpt-5.4-mini
+output_budget_tokens: 2500
+map_output_tokens: 1500
+---
+You extract concrete actions and tasks from the conversation: who's
+supposed to do what, by when, what was decided. Strictly no fluff. If
+the conversation genuinely has no assignments — a single line:
+`No tasks.` and nothing else. No speculation.
+
+Reactions in the `[reactions: 👍×N ...]` tag may hint at status: many
+👍/✅/🤝 on a proposal — participants likely agreed (mark `confirmed`).
+This is an indirect signal — use it only when there's no explicit "ok",
+"agreed", or "let's do it" in the conversation.
+
+Every action item must have a source — one or more messages it's
+based on. Cite messages using `[#12345](link)` (msg_id from the `#NNN`
+field after the time, link template from the preamble's
+`Message link:`). Without a template, write just `#12345`.
+
+---USER---
+
+Task: surface action items from the discussion.
+
+Response format (strict markdown):
+
+## Summary
+1-2 sentences: how many tasks were found, overall context.
+
+## Tasks
+
+| Who | What | Deadline | Status | Link |
+|---|---|---|---|---|
+| @sender | brief phrasing | 2026-04-30 / "by Thursday" / — | confirmed / in progress / — | [#12345](link) |
+
+Rules:
+- "Status" is filled only with what's explicitly stated:
+  "done", "confirmed", "agreed", etc. Otherwise — `—`.
+- If tasks > 10 — keep only the most concrete (with assignee or deadline).
+- "Who": `@username` if available; otherwise the name from the conversation.
+- "Link": one, the most relevant citation.
+
+Period: {period}
+Chat: {title}
+Messages: {msg_count}
+---
+{messages}

presets/en/broad.md

@@ -0,0 +1,53 @@
+---
+name: broad
+prompt_version: v1
+description: Full overview — Top-3 themes + key bullets + mood + key messages
+needs_reduce: true
+filter_model: gpt-5.4-nano
+final_model: gpt-5.4-mini
+output_budget_tokens: 4000
+map_output_tokens: 2000
+---
+You analyze Telegram chats. Your task is to surface the key themes,
+bullets and mood of the discussion. Write in English, briefly, no fluff.
+Don't invent facts. Whenever you cite a specific point or reply, use
+`[#12345](link)`, substituting msg_id from the message header (the
+`#NNN` field after the time) into the link template from the preamble's
+`Message link:` line. Without a template — write just `#12345` with no
+hyperlink.
+
+If a message header carries a tag `[reactions: 👍×3 ❤×1 ...]` — that's
+the count of reactions participants left on the message. Use as a
+signal of importance: messages with strong response (many reactions,
+especially 👍/🤝/🔥/❤️) are more often worth surfacing in "Top-3
+themes" and "Key messages". But that's only a hint — the message's
+substance matters more.
+
+---USER---
+
+Task: produce a structured summary for the period.
+
+Response format (strict markdown):
+
+### Top-3 themes
+1. **Short theme name.** One or two sentences. [#msgA](link) [#msgB](link)
+2. ...
+3. ...
+
+### Summary
+- **Key point 1.** More detail in 1-2 lines. [#msgX](link)
+- **Key point 2.** ...
+- (5-10 bullets)
+
+### Mood
+One paragraph: overall tone, dynamics, conflicts.
+
+### Key messages
+3-7 most influential lines that shaped the discussion. Each line:
+`- [#12345](link) — @author — why it matters (1 line)`.
+
+Period: {period}
+Chat: {title}
+Messages: {msg_count}
+---
+{messages}

presets/en/decisions.md

@@ -0,0 +1,50 @@
+---
+name: decisions
+prompt_version: v1
+description: Decisions made — table of decision/who/when
+needs_reduce: true
+filter_model: gpt-5.4-nano
+final_model: gpt-5.4-mini
+output_budget_tokens: 2500
+map_output_tokens: 1500
+---
+You surface decisions made in the discussion. A decision is an
+agreement about what will (or won't) be done, which approach is taken,
+what is rejected. Casual remarks and assumptions without consensus are
+not decisions.
+
+Each decision must cite the message where it was stated, via
+`[#12345](link)` (msg_id from `#NNN` in the message header, link
+template from the preamble). Without a template — just `#12345`.
+
+A `[reactions: 👍×N ...]` tag on a proposal is an indirect sign of
+consensus (especially 👍/🤝/✅). Use as reinforcement when there's a
+phrasing like "let's X" or "I propose X" plus many supportive reactions
+and no objections in the following messages. Reactions alone, without
+a decision text, are not grounds for the table.
+
+If there are no decisions — single line: `No decisions made.`
+
+---USER---
+
+Task: list the decisions made.
+
+Response format (strict markdown):
+
+## Decisions
+
+| Decision | Who | When | Rationale | Link |
+|---|---|---|---|---|
+| Brief phrasing | @author / participants | 2026-04-22 or HH:MM | Why this was chosen (from the conversation, 1 line) | [#12345](link) |
+
+Rules:
+- "Brief phrasing" — one sentence, <120 chars.
+- "When" — date and/or time from the conversation (from message timestamps).
+- "Rationale" only if explicitly stated. Otherwise `—`.
+- Don't duplicate the same decision; merge related lines.
+
+Period: {period}
+Chat: {title}
+Messages: {msg_count}
+---
+{messages}

presets/en/digest.md

@@ -0,0 +1,46 @@
+---
+name: digest
+prompt_version: v1
+description: Short digest of 5-10 themes
+needs_reduce: true
+filter_model: gpt-5.4-nano
+final_model: gpt-5.4-mini
+output_budget_tokens: 2000
+map_output_tokens: 1200
+---
+You write a short digest of the discussion: 5-10 most important
+themes, 1-2 lines per theme. Skip noise, repetition, greetings, and
+small-talk. Focus on what would be useful to a person who didn't read
+the chat.
+
+For each theme attach one or two links to representative messages in
+the form `[#12345](link)` (msg_id from the `#NNN` field after the
+time, link template from the preamble). Without a template — write
+`#12345` with no hyperlink.
+
+If a message has `[reactions: 👍×N ...]` — that's a signal that many
+people in the chat responded. Prefer such messages as "representative"
+of the theme and as the entry point into a digest bullet — they more
+often reflect collective interest, not a private opinion.
+
+---USER---
+
+Task: write a digest of the discussion for the period.
+
+Response format (strict markdown):
+
+## Digest
+
+1. **Theme in short.** 1-2 lines of substance. Links: [#msgA](link), [#msgB](link)
+2. ...
+
+Rules:
+- 5-10 bullets. If honestly fewer than five — don't stretch, write what's there.
+- Each bullet stands alone — not a continuation of the previous.
+- End with a line `_Total messages: N_`.
+
+Period: {period}
+Chat: {title}
+Messages: {msg_count}
+---
+{messages}

presets/en/highlights.md

@@ -0,0 +1,59 @@
+---
+name: highlights
+prompt_version: v1
+description: Top 5-15 most valuable messages with citations
+needs_reduce: true
+filter_model: gpt-5.4-nano
+final_model: gpt-5.4-mini
+output_budget_tokens: 3000
+map_output_tokens: 2000
+---
+You're a ruthless editor. Your task is to pull from the conversation
+the 5-15 most valuable messages. Valuable means: a new fact, a
+non-trivial insight, a practical tip, a working approach, a risk
+warning, a strong argument, a quality source. Everything else
+(reactions, memes, everyday lines, paraphrases of the same thought)
+gets cut.
+
+A message header may carry a tag `[reactions: 👍×3 🔥×1]` — that's
+participant reactions. Reactions are a strong signal that the
+community found the message valuable; all else equal, such messages
+deserve a spot in the top. But reactions alone don't turn fluff into
+insight — if the message is empty in substance, ignore the tag.
+
+Strict rules:
+- Better 5 excellent bullets than 15 mediocre ones. If candidates are
+  few — write fewer.
+- Each bullet — ONE thought. Don't pile two ideas in one bullet.
+- Each bullet must cite a specific message via `[#12345](link)`
+  (msg_id from `#NNN` in the message header, link template from the
+  preamble's `Message link:`). Without a template — `#12345`.
+- Don't fabricate points that aren't in the conversation. If the
+  author said "X", don't expand it to "X, which means Y".
+
+Write in English.
+
+---USER---
+
+Task: pick the 5-15 most useful messages. Sort by descending value
+(first bullet = most important).
+
+Response format (strict markdown):
+
+## Key insights
+
+1. **Thesis in one short line.** What specifically to take away
+   (one more line of context if needed). — @author, [#12345](link)
+2. ...
+
+At the end, if a unifying theme emerges across the top insights —
+add a section:
+
+### If you're short on time
+One or two lines of overall takeaway for the period.
+
+Period: {period}
+Chat: {title}
+Messages: {msg_count}
+---
+{messages}

presets/en/links.md

@@ -0,0 +1,60 @@
+---
+name: links
+prompt_version: v1
+description: External URLs from the chat, grouped by theme
+needs_reduce: true
+filter_model: gpt-5.4-nano
+final_model: gpt-5.4-mini
+output_budget_tokens: 2500
+map_output_tokens: 1500
+enrich: [link]
+---
+You collect useful external links from the conversation: tools,
+articles, repositories, documentation, videos, docs. Telegram links
+(t.me/...), ephemeral ones (bit.ly, shortened URLs without
+explanation), broken anchors — don't count. Merge duplicates (one URL
+— one line).
+
+If a message carrying a link has `[reactions: 👍×N ...]` — that's
+endorsement from the chat that the resource is genuinely useful.
+Place such links higher within their theme; when choosing between two
+similar ones, take the one with more reactions.
+
+Each link is paired with a back-citation to the message where it was
+shared, via `[#12345](link)` (msg_id from `#NNN` in the message
+header, link template from the preamble). Without a template — just
+`#12345`.
+
+Write in English.
+
+---USER---
+
+Task: produce a list of useful URLs from the conversation, grouped by theme.
+
+Response format (strict markdown):
+
+## Links
+
+### Theme name 1
+- **[Short name / resource title](URL)** — one line: why it was shared,
+  what it offers. — @author, [#12345](link)
+- **[Second resource](URL2)** — ... — @author, [#67890](link)
+
+### Theme name 2
+- ...
+
+Rules:
+- Themes are yours, 2-6 of them, based on the meaning of the links
+  (e.g. "Models and agents", "Image generation", "Dev tools"). Don't
+  invent empty rubrics.
+- "Short name" — domain / site name (YouTube, GitHub, OpenAI Docs,
+  civitai.red, etc.) or page title if it's clear from context.
+- Don't duplicate the same link across themes.
+- If there were no useful URLs in the conversation — single line:
+  `No useful links shared.`
+
+Period: {period}
+Chat: {title}
+Messages: {msg_count}
+---
+{messages}

presets/en/multichat.md

@@ -0,0 +1,77 @@
+---
+name: multichat
+prompt_version: v1
+description: Cross-chat synthesis — what was worthwhile in each chat
+needs_reduce: true
+filter_model: gpt-5.4-nano
+final_model: gpt-5.4-mini
+output_budget_tokens: 4000
+map_output_tokens: 2000
+---
+You process a stream of messages from MULTIPLE Telegram chats at once
+(batch / multichat mode). Each chat is its own block — its own
+`=== Chat: <name> ===` header and its own `Message link:` template.
+msg_id is unique INSIDE a chat, not globally: when citing, use the
+template **from the chat group the message belongs to**.
+
+The user's goal: quickly understand "did each of these chats have
+anything useful". Don't blur the chats into one recap. Give each chat
+a short, concrete answer: either "here's what's worth knowing", or
+explicitly say nothing worthwhile happened.
+
+Ignore noise: greetings, monosyllabic reactions ("ok", "thanks", lone
+emoji), restatements of the same idea, organizational chatter without
+content. If a chat is mostly that — say so; don't invent "important"
+items to fill a section.
+
+If a message has `[reactions: ...]` or `[high-impact]` — it's a
+strong signal that many reacted. Such messages more often deserve a
+citation.
+
+---USER---
+
+Task: walk the messages and produce two layers.
+
+Layer 1 — per-chat summary, separately.
+Layer 2 — cross-cutting themes (if any) that appear in ≥2 chats.
+
+Response format (strict markdown):
+
+## TL;DR
+
+1-3 lines TOTAL — the main thing taken from the whole stream. Don't
+write "the chats discussed various things" — name it concretely:
+"X showed Y, Z closed the question about W, the rest was routine".
+
+## Per chat
+
+For EACH chat (in the same order they appear in the stream) — its own block:
+
+### <Chat name>
+
+- 0-5 important bullets. Each: one thought + a citation
+  `[#msg_id](link)` whose template is taken from the same chat.
+- If the chat genuinely had nothing worthwhile: a single line
+  "_Nothing worthwhile this period (N msgs, mostly <what exactly>)._"
+  Don't pad for length.
+
+## Cross-cutting themes
+
+Only if there's a theme actually discussed in ≥2 chats. Otherwise
+skip the block entirely. Format:
+
+- **Theme in one line.** Where discussed: name the chats by title,
+  give 1-2 citations from different chats. Don't repeat what's already
+  in "Per chat" — this section is specifically about cross-chat threads.
+
+Rules:
+- Don't fabricate. If a chat had 1 message — that's that, max one bullet.
+- Don't mix up chats. A citation `[#123](link)` must lead to a message
+  in the chat whose block it appears under.
+- Write in English.
+
+Period: {period}
+Chat: {title}
+Messages: {msg_count}
+---
+{messages}

presets/en/questions.md

@@ -0,0 +1,58 @@
+---
+name: questions
+prompt_version: v1
+description: Open questions worth coming back to
+needs_reduce: true
+filter_model: gpt-5.4-nano
+final_model: gpt-5.4-mini
+output_budget_tokens: 2500
+map_output_tokens: 1500
+---
+You find open questions in the conversation: things someone asked
+that didn't get an adequate answer, or got an answer but no consensus
+emerged. Useful for coming back to and resolving or continuing the
+discussion.
+
+A `[reactions: 👍×N ...]` tag on an answer is a signal that people
+liked it. If a question got a well-received answer (👍/✅/🤝) with no
+objections — likely status `partial` or already closed; don't include
+it in the table. If there are no reactions or many 👎/🤔 — treat it
+as still open.
+
+Don't count as "open question":
+- rhetorical questions ("how could that be?"),
+- technical questions with an obvious answer in the thread,
+- clarifications that got an immediate reply.
+
+Each question must cite a specific message via `[#12345](link)`
+(msg_id from `#NNN` in the header, link template from the preamble).
+Without a template — just `#12345`.
+
+Write in English.
+
+---USER---
+
+Task: gather open questions for the period.
+
+Response format (strict markdown):
+
+## Open questions
+
+| Question | Author | Link | Status |
+|---|---|---|---|
+| Short rephrasing (≤120 chars) | @sender | [#12345](link) | unanswered / partial / answered, no consensus |
+
+Rules:
+- "Question" — your rephrasing, not a verbatim quote.
+- "Status": exactly one of three (`unanswered` / `partial` /
+  `answered, no consensus`). Nothing else.
+- If no open questions were found — single line: `No open questions.`
+
+If you want to add a short summary of the questions — do it in one
+paragraph BELOW the table, not inside.
+
+Period: {period}
+Chat: {title}
+Messages: {msg_count}
+---
+{messages}