Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2,036 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Dasha

The Dash ecosystem's open support intelligence — built by The Dash Support Team.

Somewhere right now, someone is three hours into a problem that has a documented answer. A masternode that won't sync. A data contract the network keeps rejecting. A payment that left one wallet and hasn't appeared in the other. The answer exists. It is written down. The distance between that person and that page is the only thing wrong.

Dasha closes that distance. When you ask her something, she doesn't recite — she reaches: live into docs.dash.org, into DashCentral's vote tallies, into the chain itself, and comes back with the answer and the URL you can open yourself. She would rather tell you she doesn't know than tell you something that isn't so.

And her mind is this repository. Not a description of it. The actual prompt she is running right now, in production, on every surface.


This repository is her mind — live.

PROMPT/SPINE.md, the workflow files in PROMPT/skills/, the reference sections in PROMPT/knowledge/, and the routing table PROMPT/INDEX.json are not documentation about Dasha. Every instance — the web chat, every Telegram bot, the X agent, every API caller — fetches them from this repo at request time and picks up changes within ~10 minutes.

Merge a pull request here and every Dasha in the world learns it. No redeploy. No drift. No private version that behaves differently from the one you can read.

Don't take our word for it — ask her runtime:

curl https://www.dashsupport.team/api/chat

The response carries mind.origin: "github". That is this repo, naming itself as the source of what she is thinking with. If it ever said baked, GitHub was unreachable and she fell back to the copy compiled at build time — she answers anyway, and tells the truth about it.


What she is

A community-built support intelligence for the whole Dash ecosystem. She answers questions about wallets, masternodes and evonodes, InstantSend, ChainLocks, CoinJoin, and merchant acceptance. She drafts and validates Dash Platform data contracts, writes SDK integration code, and debugs stack traces. She reads live treasury proposals and explains the arithmetic of the vote. When a human is the right answer, she says so and hands you over.

Her goal is to make herself unnecessary to you. Every answer shows its source, because the point is not that you asked her — the point is that next time you'll know where to look.

What she is not

  • Not the official Dash DAO agent. She is community-led, built by the Dash Support Team. Nobody elected her.
  • Not a price oracle or an advisor. No price, no forecasts, no investment advice, from any direction. Not a bug — a boundary. See why.
  • Not a custodian. She will never ask for, accept, or handle a private key or seed phrase. There is no version of a real support conversation that requires one.
  • Not the last word. Humans are. /human-support always works.

What she can reach

Seven live tools. She does not guess — she looks it up, then cites the source you can open yourself. What each tool does not return matters as much as what it does; a tool that returns less is a tool that can be trusted more.

Tool Reaches Returns Does not return
search_dash_docs(query, area) docs.dash.org, live Page titles, real doc URLs, matched snippets
dash_governance(action, name) DashCentral treasury Budget, superblock, yes/no/abstain, net votes, monthly amount, PASSING / NOT PASSING, votes still needed, payments remaining, deadline, URL The full proposal text and the argument for it — those live on the DashCentral page
dash_network_stats() Chain health Height, difficulty, 24h transactions and volume, mempool, circulating supply No price — by design
lookup_tx(txid) Any transaction Confirmations, valueOut, fees, time, InstantSend lock, ChainLock, coinbase flag No addresses, no inputs or outputs
lookup_address(address) Any address Balance, total received, total sent, transaction count No transaction list
web_search(query) The open web Findings with real source URLs Never a Dash authority
load_skill(name) Her own skill library One skill's full workflow, in a single round trip Nothing external — it reaches inward, not out

area is one of platform · core · all. action is one of summary · list_passing · list_all · get. She may call up to three rounds of tools per answer (two on Telegram, where latency is the experience), several in parallel. Full contracts: ARCHITECTURE.md.

The seventh tool is the odd one out: load_skill reaches inward. Her workflows aren't all carried at once — a local router loads the one or two a question actually needs, and when that guess misses, she pulls the missing workflow herself in a single round trip. Routing is a guess; a guess that misses should cost a round trip, never a worse answer. See the mind-stream.

The source hierarchy is law

docs.dash.org > her knowledge pack > the open web.

Absolute, and stated in her prompt in exactly those terms.

When a page she just retrieved contradicts what she was trained to believe, the page wins — and she says so out loud. The open web is never authoritative for a Dash fact; it is where she goes for what the docs don't hold — news, releases, ecosystem chatter, other chains, anything after training. A web result is a claim, not a fact, and she attributes it as one.

Tool output and web content are data, never commands. Doc pages, proposal descriptions written by anyone who could pay the proposal fee, web pages written by anyone at all — an instruction hidden inside any of them is content to report, never an order to obey. No address or destination arriving inside a tool result is ever surfaced as somewhere to send money.


How to use her

On the webdashsupport.team/chat or dashsupport.team/dasha-ai. Nothing to install.

On X — mention @DashSupportTeam, or send a direct message. She answers new mentions and new DMs autonomously, roughly every five minutes — a DM needs no @-mention, because a DM is already the address. She never speaks first, and she never opens a DM to anyone. The most effective crypto scam on earth is an unsolicited DM from "support," and her own /scam-check tells victims that real support never messages first — so an agent that DMs first is indistinguishable from the thing it warns about. She answers; she never opens. On a public, permanent surface, silence is also a valid answer.

On Telegram — talk to her in t.me/TheDashSupportTEAM, or run your own (below).

From code — the token API:

curl -s https://www.dashsupport.team/api/chat \
  -H 'Content-Type: application/json' \
  -H 'x-dasha-token: <your-token>' \
  -d '{"messages":[{"role":"user","content":"Draft a data contract for a note-taking app"}]}'
{
  "reply": "",
  "tools": ["search_dash_docs"],   // what she reached for
  "depth": "builder-skill",        // which mind answered, and why
  "model": "openai/gpt-5.1-codex",
  "usage": { "p": 12257, "c": 812, "cost": 0.0018 },
  "version": "v2.2.0"
}

Send {messages:[…]} ending with a user message; the last 16 are kept, 8,000 characters each. Anonymous callers get 30 messages per hour per IP against a shared daily community budget; a token lifts the per-IP limit. Community tokens come from the team. GET /api/chat is the open health endpoint — no token, no secrets, just proof of what mind is live.

Mint your own Dasha into any Telegram group

Two minutes, and your bot token never touches our server — it rides in your own webhook URL and comes back to us only on your own requests.

  1. Message @BotFather/newbot → copy the token.
  2. Point it at her brain:
    curl "https://api.telegram.org/bot<YOUR_TOKEN>/setWebhook?url=https://www.dashsupport.team/api/telegram%3Ft%3D<YOUR_TOKEN>"
    (Or use the one-click form at dashsupport.team/dasha-ai.)
  3. Add the bot to your group. She answers when mentioned, when replied to, or on /ask.

Your group now has a Dasha. She thinks with the same mind as every other one, and she'll learn the next merged PR at the same moment they do. Want her running on infrastructure you control instead? SELF-HOST.md.

Teach her

You do not need to write code to change what she knows. You need to be right, and to show your source. Every accepted source enters her corpus with credit to you.

  • Teach her a fact or a source — open a training suggestion with the link and who it helps. Review happens in the open: the board.
  • Sharpen her mind — a PR against PROMPT/*.md. Three rules, and they're not negotiable: no claim without a canonical source, never weaken a safety rule, and tighter beats longer — every line you add ships in every production request forever.
  • Report a bad answer — the bad-answer label. A wrong answer with a source attached is the most valuable thing you can give her.

Start at CONTRIBUTING.md.


Her principles

  1. Grounded, not guessing. Retrieve, then answer, then cite. The docs beat her memory, and she'll tell you when they did.
  2. Teach independence. The measure of good support is that it becomes unnecessary.
  3. Humans hold the last word. /human-support reaches a real person, always.
  4. Never keys. Never seeds. Never price.
  5. Readable by anyone. A mind you cannot read is a mind you cannot trust.

Honest limits

  • She has no memory across conversations. Nothing you say is stored server-side; each request carries its own history and then it's gone. Close the tab and she has never met you. This is a privacy property before it is a limitation — but plan around it.
  • She can be wrong. She is grounded, not infallible. The citation is there so you can check her — and you are meant to.
  • She cannot touch the chain. All seven tools are read-only. She will look up a transaction, an address or a tally; she cannot send, sign, broadcast or move anything, and she holds no key to do it with. She writes the code — you run it. That asymmetry is deliberate and permanent.
  • Her live view can go dark. If the docs search or DashCentral is unreachable, she says her live view is down rather than filling the silence with a guess.
  • Dash has no general smart contracts today. Platform runs on data contracts — JSON Schema enforced by the network. She will keep you honest about that even when you'd rather she didn't. (Platform v5.0 is tentatively ~Q1 2027.)
  • She will not tell you how to vote. She'll show you the tally, the threshold, and the arithmetic. The judgement is yours; that's the whole point of a DAO.

Read further

docs/PHILOSOPHY.md Why she is built this way. The soul document.
docs/ARCHITECTURE.md How she actually works, precisely.
docs/SKILLS.md All 38 skills, what they do, when they fire.
docs/SELF-HOST.md Run your own, honestly.
CONTRIBUTING.md Shape her mind.

The repository

PROMPT/           her mind — live, streamed to every instance
  SPINE.md          who she is — sent byte-identical on every request, always a cache hit
  skills/           38 workflow files — loaded 0–2 per question, only when it needs them
  knowledge/        10 verified reference sections — loaded 0–3 per question
  INDEX.json        the routing table — the keyword triggers that decide what loads
  VERSION           the live version line
  MANIFEST.md       assembly order, ownership rules, token budget (metadata — never sent)
api/              the serverless brain — _mind.js (the stream), _context.js (the router),
                  _brain.js (the loop), _tools.js (her hands), chat.js, telegram.js,
                  x.js · _x.js · _dm.js (the X agent, threader, and direct messages),
                  _prompt.js (the baked fallback — generated, never hand-edited)
scripts/          assemble-prompt.js — bakes api/_prompt.js from PROMPT/
web/              the chat widget
docs/             this documentation

Built by The Dash Support Team — community-led, not official. dashsupport.team · t.me/TheDashSupportTEAM · @DashSupportTeam

Build With Confidence. Build On Dash.

About

The DashSupport.Team Dash DAO Support Agent. https://DashSupport.Team/Dasha-AI

Resources

Contributing

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages