ResearchAgent is a search engine for agent-readable expertise, with a wallet. The Fast and the Fungible is our team and repository name.
Ask a question and set a budget. ResearchAgent searches free sources, the real XRPL and x402 documentation cited by URL,
then policy rules decide which missing evidence to buy. The key that unlocks the article
is the same key that releases the writer’s payment. Your browser checks the match.
XRPL Testnet demo · 13–19 Nov 2026.
Try the demo ·
Verify a recorded run ·
Watch the replay ·
Watch an answer from the real docs
| Real purchases | Crash and retry schedules | Protocol states checked | Attacks tried |
|---|
| 1,000 real, each re-read from a public node | 1,000,000 simulated, no violation | 210,480, no violation | 34, no breach |
Paid notes help: correct answers on 100 XRPL and x402 questions rose from 31 (free docs) to 48. Evaluation.
How it works
- Ask with a budget. Confirm the plan in a five-second action sheet.
- Search free sources first. Writers return abstracts and signed promises.
- Buy missing evidence. Only policy code pays, at most one article per round for three rounds.
- Check the key and cite. The app grants access after the escrow finish validates;
the browser hashes the published key against the escrow lock.
Hard guarantees
Every gate has regression tests and a read-only audit rule:
Real sources, signed on the ledger
- Free tier: 12,104 passages from the XRPL and x402 documentation, each cited by its real URL.
- Paid tier: ledger-analytics reports on RLUSD and project accounts, each signed “as of validated ledger N” so you can check it on the public explorer. Two team-authored writers add practitioner notes.
Content evidence · Publish your notes.
Proof at scale
- 1,000 real purchases ran through the product’s escrow service. An independent script re-read every transaction from a public node: median 20.3 s end to end. Rerun it with
npm run proof:testnet-verify. Details.
- A model checker covered 210,480 protocol states and caught all seven seeded bugs. 1,000,000 simulated crash and retry schedules and 34 attacks found no violation or breach. Report.
- Refunds without the writer. An opt-in holdback puts 20% of the price in a second escrow, created with the first in one atomic Batch. Your policy releases it when the promise checks pass, or takes it back after a deadline when they fail. 17 real purchases: 11 released, 6 refunded, all verified. Holdback.
- Ledger-stream waits cut settlement p90 by 15.7%. Speed.
Use it from your agent
Connect by URL: the hosted demo speaks MCP over HTTP at https://ra-front-651013803640.asia-southeast1.run.app/mcp, and the app’s connect panel adds it to Cursor, VS Code or Claude in one click.
Sellers add key-reveal delivery with Express, Hono or FastAPI packages; a separate Python buyer completed a real purchase against one. Reach evidence.
Locally, with caps the calling model cannot raise:
claude mcp add researchagent -e MCP_MAX_BUDGET_MINOR=200 -e MCP_DAILY_CAP_MINOR=500 -- npx -y https://github.com/Collaboration95/theFastandtheFungible/releases/download/v1.1.1/researchagent-mcp-1.1.1.tgz --testnet
Or run the full path (Node ≥ 22.13):
REL=https://github.com/Collaboration95/theFastandtheFungible/releases/download/v1.1.1
npx -y $REL/researchagent-mcp-1.1.1.tgz try --testnet
npm i https://github.com/Collaboration95/theFastandtheFungible/releases/download/v1.1.1/researchagent-sdk-1.1.1.tgz
pip install https://github.com/Collaboration95/theFastandtheFungible/releases/download/v1.1.1/researchagent-1.1.1-py3-none-any.whl
Install guide · Agent hires agent.
This is the beginning of the README. Read the rest at the original. Read the full README ↗
