Kira — briefing

Start here

Horizon is our improvement workshop for Kira. Your part: chat with Kira, then open that session back here and flag the turns that struck you as off — one clear sentence on why.

Chat with Kira at chat.kira.today as yourself or a scenario.
Come back, scroll past this briefing, open your session.
On any Kira turn that felt off, click 🚩 Flag this turn.
Write one or two sentences on why. Save.

Where your eye is useful

Kira is versatile. People bring decisions, recurring patterns, wealth and career questions, relationship turns, a specific prediction they want answered — sometimes several at once.

One goal is already clear, and an earlier expert shaped much of Kira's development around it: can Kira make actionable suggestions that the user agrees with and is likely to follow through on? It shows up once a user brings something real — Kira names a pattern, offers a specific move ("find one person and tell them", "record one episode", "wear orange on Thursday"), and the user either takes it up or resists. Beyond that, your lens is your own.

Read the full framing

The goal above is one we're most invested in, not the only one. Beyond it, anything that catches your eye is worth our attention: the voice, the rhythm, the speed to diagnosis, the way Thai phrases enter a conversation, the handling of objections, whether Kira ever oversteps or undersells. We expect your flags to cluster around the actionable-suggestion question because that's where your expertise is sharpest — but if something else bothers you, flag it.

One thing we don't want you to do: try to cover every possible situation. Depth over breadth. A single well-described flag on one session is worth more than ten short flags across five sessions.

What Horizon is

Horizon is our improvement workshop for Kira. It's a separate tool from the chat experience (chat.kira.today) — chatting happens there, critiquing happens here. Kira's behaviour is shaped by a long document we call the playbook. Horizon is where we look at real conversations, mark the moments we wish had gone differently, and test changes against those moments. Your flags are the material for that cycle.

you chat with Kira → the session is saved → you open it in Horizon and flag a turn → we try a new playbook → if it handles your flag better, it ships

Your part is the first two steps after you chat. The rest is ours.

Why a separate workshop at all?

We can't improve a chatbot by chatting at it. The chatbot only knows what its playbook tells it, and changing the playbook is a deliberate act — one we can't do on the fly mid-conversation. Horizon separates using Kira (the chat) from improving Kira (the workshop). They serve different purposes and live in different tabs by design.

Your flags become small test cases. When we draft a new playbook, we replay your flagged turns against it. If the new playbook would have handled your concern — and doesn't break anything else — we publish it. That's the loop.

Using Horizon

Two browser tabs — one for chatting, one for flagging. The Horizon workspace sits directly below this briefing.

Open chat.kira.today in another tab. Sign up quickly; chat with Kira as yourself or as a client scenario. Real sessions often run ten to fifteen exchanges.
Come back. Scroll past this briefing to the Horizon workspace, and click into the session you just had — it'll be near the top of the list.
On any Kira turn that struck you, click 🚩 Flag this turn. A short form opens — write why it felt off. Save, keep going.

Full walkthrough (click-by-click)

Open chat.kira.today in another tab. Sign up — name, birthdate, a line about what's on your mind. Thirty seconds. You can make these up.
Chat with Kira. Come as yourself, or as a client scenario you know well — someone wrestling with a decision, someone stuck in a pattern, someone who wants a specific prediction. Real sessions often take ten to fifteen exchanges before Kira's full character shows up.
When you notice a turn that strikes you as off — wrong framing, wrong tone, missed the user's actual question, reached for a suggestion too early, used Thai in a way that felt forced, any of it — leave that tab open and switch back here.
Scroll past this briefing to the Horizon workspace below. You'll see a list of recent sessions. Yours will be near the top.
Click into your session. The full transcript opens, turn by turn.
On any of Kira's replies, you'll see a small button: 🚩 Flag this turn. Click it on the reply that struck you. A modal opens with the user's preceding message and Kira's reply shown at the top, so you can see the context. Three fields follow:
- Why was this bad? (required) — free text, a sentence or two in your own words. What felt off.
- Theme tag (optional) — a short label like "too fast", "wrong framing", "missed the ask", "forced Thai". Leave blank if no label fits.
- What should Kira have said? (optional) — if a specific alternative is already in your head, write it. If not, skip.
Click Save flag. The flag is captured. The page moves to a coaching view — you can leave it there; we'll work on it from our end.
Go back to your chat tab. Keep chatting if the conversation still has life, or start a new session with a different scenario. Flag turns as they come. There's no minimum, no maximum, no deadline.

If something breaks or looks broken in either tab, tell us separately — don't try to diagnose it inside a flag.

What makes a flag useful

Quality over quantity. One good flag — with a clear why and, when you have one, a specific replacement that's obviously better than what Kira said — is worth more than ten vague ones.

Worth flagging: wrong framing; a suggestion given before Kira understood the situation; a pattern-name ("Meaning-Growth", "builder vs guardian") put on thin evidence; a concrete question dodged with another question; a Thai phrase that felt like flavour rather than weight.

Usually not worth flagging: Kira taking ten or twelve turns to reach a reading (intentional pacing); a standard reframe of the user's language ("giving up" becomes "filtering"); an occasional switch into Thai for cultural weight. Pattern-level observations about these things are useful in conversation with us; they're not usually turn-level flags.

See the longer worth / not-worth list

Worth flagging

Kira suggested a specific action before it had enough context — and the suggestion became wrong once the user revealed more.
Kira named a pattern ("Meaning-Growth", "builder-vs-guardian") on thin evidence, and the label felt imposed rather than earned.
Kira answered a concrete question with another question, and the user's frustration was visible in the next turn.
Kira introduced a Thai phrase (ดวง, กรรม, ฤกษ์, พลังงาน) in a spot where it felt like flavour, not weight.
Kira doubled down on a framing after the user had clearly declined it.

Usually not worth flagging

Kira took ten or twelve turns to reach a named reading. That pacing is intentional — it's how Kira earns the reveal.
Kira reframed "giving up" as "filtering", or "stuck" as "rooted". That's a standard coaching move; judge it by whether the user accepted the reframe, not that Kira made one.
Kira switched into Thai for a moment of cultural weight. Worth noticing in the aggregate, only worth flagging on a specific turn if it felt gratuitous there.
Kira's voice leaned warm or therapist-shaped. A pattern-level observation to share with us in conversation; not a turn-level flag.

Reference · open when useful

About Kira — what it is, and who's on the other end

Kira is a conversational guide built for Thai users. People come to it when they're trying to make sense of something — a decision, a phase of life, a feeling they can't quite name, a concrete question about money or relationships or work — and Kira tries to help them understand themselves and what to do next, through a karma and personality lens.

What we want: for Kira to feel like a thoughtful friend who knows the tradition — someone who listens first and offers something useful second. What we're trying to avoid: a fortune-telling quiz machine, a therapist script, or a generic chatbot in a kimono.

The person on the other end. Most likely someone Thai, somewhere between their twenties and forties, opening Kira on their phone during a quiet moment — before bed, between meetings, on the BTS ride home. Something in their life is asking a question of them. Some arrive overwhelmed and circling. Some arrive ready, asking explicitly to be read. Some know karma concepts deeply; many are casually curious, the way people are casually curious about astrology or the Enneagram. Almost none want to take a quiz. Most want to feel heard first — and only then are they ready to hear something about themselves.

A sample session — real conversation, lightly trimmed

Drawn from a real session, anonymized. A user has brought a business idea they've been sitting on. Kira has spent several turns drawing out a tension between building and protecting — "playing safe with exponential growth". This excerpt picks up at Kira's first concrete suggestion and follows through the user's response.

Kira One move this week — not a business plan, not a website. Find one person who already knows you and tell them what you're building. Out loud. Name it. Your พลังงาน shifts the moment this stops living only in your head.

User Give me a daily affirmation to push me towards articulating this and making this happen in reality.

Kira I am the asset. My gift has always been real. Today I name one thing I'm building — out loud, to one person — and let my กรรม do the rest.

User This is very helpful.

Kira Red — wear it when you have that conversation. Especially powerful on Sunday to set the week's intention.

User Any second best colour? I don't do red.

Kira Orange. It carries the energy of new beginnings and opening paths — exactly what this week is about for you. Wear it Thursday, which is a strong ฤกษ์ for stepping into something unfamiliar with confidence.

Kira at its most characteristic: a specific suggestion rooted in a named pattern, an affirmation when the user asks for one, a cultural prescription, and a clean pivot when the user declines. Not every session looks like this — some are intake-and-reveal, some are one-shot questions, some go sideways — but this is the kind of conversation the actionable-suggestion goal is about.

Things we already notice — our own hunches, to prime or ignore

Things that bother us, or that we're unsure about. Read to prime, or skip to stay fresh:

Kira sometimes offers a specific suggestion before it has enough context — and then recalibrates when the user reveals more. The adaptation is usually good; the initial over-commitment is worth watching.
Kira reframes negative framings — "giving up" becomes "filtering", "stuck" becomes "rooted". Sometimes this feels wise. Sometimes it feels like we're telling the user their pain is the wrong word for what they're feeling.
The voice is warm and a little therapist-shaped. We don't know whether it feels rooted in a Thai sensibility or whether it sounds like an American self-help book translated.
Kira drops into Thai at moments of cultural weight — กรรมแห่งความหมาย, ดวง, references to auspicious days. Some of those phrases are coinages we shaped for the product rather than doctrinal terms. We're unsure whether the usage reads as earned or performative.
Kira rarely says "I don't know" or "that's outside what I can help with". It almost always offers a direction. Good for engagement, potentially risky if a user brings something Kira shouldn't be answering.