> ## Documentation Index
> Fetch the complete documentation index at: https://cofounder.appeeky.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Chat with your Co-founder (Ask)

> A full conversation with your co-founder — it reads your live data, remembers past decisions, and can approve actions for you.

Ask is a real chat with your co-founder, not a one-shot Q\&A. It's available in four places with the same brain behind all of them:

* **Dashboard → Co-founder → Chat** — scoped to one app, full history
* **Slack** — mention the bot or reply in a briefing thread
* **Telegram** — just message the bot
* **API** — `POST /v1/cofounder/ask` for your own tooling

## What makes it different from a chatbot

Every answer is grounded in *your* data. The agent has a tool belt it can use mid-conversation:

* **Live data tools** — keyword ranks, competitor lookups, ASO scores, App Store Connect metrics, reviews, Apple Search Ads, RevenueCat, Google Play analytics. If the answer needs fresh data, the agent fetches it during the turn.
* **My Apps workspace** — same data as the dashboard tabs: localization drafts (`list_localizations`, `get_localization`, `publish_localization`), release drafts (`my_apps_versions_*`), competitor timelines, saved reports, similar-app discovery. See [My Apps workspace](./my-apps-workspace).
* **Growth (Reddit) read-only** — inbox opportunities, projects, and scheduled posts (`growth_*` tools). Co-founder can read and summarize; posting/replying stays in the dashboard.
* **Semantic memory search** — "did we already try changing the screenshots?" works even when the wording doesn't match, because the agent searches past episodes by meaning.
* **Your company context and Library** — profiles and `business_context.md` are in every answer's context; other [Library](./library) documents are pulled in by relevance to the question you just asked. It can also search and read them explicitly (`search_documents`, `read_document`) and write notes into `research/`. Matching is cross-lingual, so a Turkish question finds an English document.
* **Context proposals** — tell it something durable ("we stopped targeting Brazil") and it files a [knowledge patch](./knowledge-patches) instead of letting it decay as a fact. It can't edit your context itself.
* **Action tools** — the agent can list your pending action queue, and (when you tell it to) approve or reject a proposed action right from the chat. This is how Telegram/Slack approvals work under the hood.
* **The latest briefing** — always in context, so "why did you recommend that?" has an answer.

When ASC live writes fail (permissions, Google Play app), write actions save to **My Apps drafts** instead of erroring — the agent should tell you to review in Localizations or Releases.

A turn is capped at 5 tool steps — enough for "resolve app → check competitors → check keywords → answer", not enough for the model to wander.

## Multi-turn conversations

Chat remembers the last \~6 turns per channel, so follow-ups work naturally:

```
You:  How's the Mexico funnel doing this week?
Bot:  Mexico is still 25% of downloads with no localization — same gap I
      flagged Monday. Conversion from MX impressions is 0.8% vs 2.1% for US.

You:  ok, what would localizing cost me in effort?
Bot:  For MX you only need es-MX metadata — title/subtitle/keywords (~1h).
      Screenshots can stay English for a first pass...
```

## What it can and can't do

| Can                                                   | Can't                                                           |
| ----------------------------------------------------- | --------------------------------------------------------------- |
| Fetch live keyword/competitor/ASC/ads data mid-answer | Run a full collector→reasoner pipeline (trigger a run for that) |
| Approve / reject proposed actions when you ask it to  | Execute *new* write actions that weren't proposed by a run      |
| Recall past episodes and facts semantically           | Access another user's data (always scoped to your account)      |
| Answer in your briefing language and persona voice    | —                                                               |

## Latency and cost

Simple questions answered from the briefing come back in a few seconds. Questions that need live data (tool calls) typically take 15–60 seconds depending on how much the agent has to fetch. Cost depends on your [model settings](../concepts/llm-settings) — with default models a typical turn is a fraction of a cent.

## Persistence

Every Q\&A is saved in your history with question, answer, source channel, and cost:

```bash theme={"theme":{"light":"github-light","dark":"github-dark"}}
GET /v1/cofounder/ask?appId=6479581124&limit=20
```

See the [Ask API reference](../api/ask) for full request/response shapes.

## Good questions to ask

* *"What changed since the last briefing?"*
* *"Why did downloads drop yesterday?"* — for deeper digs, use the [Why-agent](./why-agent)
* *"What 3 keywords should I go after this week?"*
* *"What's pending in my action queue?"* → *"approve the first one"*
* *"Was my last subtitle change worth it?"* — pulls from A/B Test Memory
* *"What did we learn from the last pricing experiment?"*
* *"Any localization drafts I need to publish?"* — [My Apps workspace](./my-apps-workspace)
* *"What did competitors change recently?"* — competitor timeline tools
* *"Show Reddit opportunities in the inbox"* — Growth read tools (no auto-post from chat)
* *"How's Google Play doing vs last month?"* — `google_play_analytics_overview`
* *"What do you know about us?"* — renders your [company context](../concepts/knowledge-base) verbatim, so you can see what it's actually working from
* *"What did I write about our paywall tests?"* — searches your [Library](./library)