Skip to the section you want, or ctrl-F. If you can't find your answer, email support@askkira.com — we read every one.
It captures everything that makes your expertise distinctive — documents, frameworks, the way you'd never approach a problem, your real voice from interviews — and synthesises it into a structured product spec that an LLM can act on. The output is a long-form system prompt, a values list, and a hand-off bundle ready for deployment.
Most "build your own GPT" tools paste your documents into context and call it a day. We turn knowledge into operating behaviour: identity, routing rules, mandatory response structures, safeguards, anti-pattern detection. Your AI sounds like you, not like ChatGPT.
20-40 minutes for a usable first build. A genuinely distinctive tool — what we call Talismanic on the confidence dial — typically takes 2-3 sittings spread over a week, because the highest-value inputs (the Socratic interview, your anti-patterns) deserve thinking time.
Top of the dashboard, fill in the setup card: tool name, audience, sector, tagline. That's the anchor for everything else. Then click the Socratic interview channel — even six honest answers raise your confidence band noticeably.
Each channel is a different way to feed your AI distinctiveness: documents (PDFs, slides), web (your site, blog), mailbox (sent folder — your real voice), cloud drive, audio/video (talks, podcasts — auto-transcribed), published works (books, papers), Socratic interview (the deepest source), frameworks (your proprietary methods), case studies (anonymised concrete examples), anti-patterns (what you'd never do — the single best values driver), and brand (logo, colours, voice).
No. Nothing you upload is used to train any model that benefits another customer. Your corpus is a private workspace, not training fuel. We don't aggregate, sample, fine-tune on, or share it. Full policy in Security & trust.
Never. Every project is scoped to your account by owner_id. Every read, write, build and download endpoint verifies that the current session belongs to the project's owner before doing anything. Project IDs are also unguessable (128-bit random) so enumeration isn't possible.
The project file and every uploaded document is wiped immediately from disk. No soft-delete, no shadow copies. If you'd downloaded the hand-off bundle, that's the only retained record.
Yes — designed that way. The hand-off bundle is a single ZIP containing the system prompt (works with any LLM that accepts a system message), your values list, the full project JSON (re-importable), and every source document. No proprietary format, no lock-in. We earn renewals, not lock-in.
On staging, files live on the server's encrypted disk under data/projects/. In production we'll move project metadata to encrypted Postgres and source documents to encrypted object storage. Either way: at-rest encryption, region-pinned, with backups.
Stored only as a pbkdf2-sha256 hash with 250,000 iterations. We can't read it; we can only check whether a submitted password matches. If you forget it, you'll need to reset (password reset comes online with the SMTP integration).
How distinctive your AI will feel. Bands: Generic (0-29) — sounds like any chatbot, don't ship. Workable (30-59) — knows your basics, useful but unspecial. Distinctive (60-84) — your voice is showing, ready to share with stakeholders. Talismanic (85+) — feels like only you could have built it.
The Socratic interview and your anti-patterns. People underestimate anti-patterns ("what we'd never do") — they're the single most powerful values driver because they encode negative space that's hard to derive from documents.
PDF, DOCX, PPTX, TXT, MD, CSV. Max 20 MB per file, 80 files per upload request. Total per project: as much as you like — we'll let you know if any single file is too big.
The upload endpoint accepts the request, saves the files under the limit, and returns a rejected list with the reason ("too large"). You'll see this in the documents channel; nothing else fails.
Yes. Use the picker in the top bar to switch between projects, or click ＋ New tool… to start a fresh one. The trial plan supports 3 projects; paid plans go higher.
You can — the bundle ZIP contains system_prompt.txt as plain text. Edit it freely, then hand it to your AI team. AI Builder re-synthesises from the channel inputs each time you press Build, so manual edits to the saved prompt will be overwritten if you rebuild.
Today: click Mark ready for hand-off, download the bundle, and send it to your AI team (or to askKira to host on your behalf). The bundle works with any LLM that accepts a system prompt. Soon: a one-click "publish to askKira" button that pushes the bundle and stands up a branded interface for your customers.
Once the askKira deployment integration ships, yes — your tool gets a branded subdomain or sits behind your paywall. While that's being built, the hand-off bundle is the canonical artefact.
That's where we're heading. Plans on the roadmap: paywall integration, per-seat licensing, revenue-share for distribution through the askKira marketplace. None of these are live yet on staging.
Yes — go to Account. You'll need your current password to set a new one (10 character minimum).
Self-serve password reset comes online with the SMTP integration. Until then, email support@askkira.com and we'll reset it manually within a working day.
From Account → Sign out, then email us to close the account. Your projects are wiped on close; if you want to keep them, download the hand-off bundles first.
Yes — every new account starts on the Trial plan with 3 projects, no card required. Upgrade when you outgrow it.