Go straight to di content

Technical details, legal documents and some account steps still dey for English.

MITHRIL / WHITE HAT RESEARCH

Empower those who protect.

Bring our strengths together. Bring light back to the world.

Some notice the dangers others overlook.

Some look deeper into systems, determined to make them better.

Some use that knowledge to protect everyday lives.

mithril.fund supports the curiosity and creativity of white hats.

One person’s discovery becomes a community’s strength.

Together, that strength can change the world.

Investigate authorized code. Validate the finding. Ship the fix. Specialized LLM inference for verified white-hat researchers, with a free research allowance.

Preparing for launch — the selected model returned a real response on the dedicated deployment. Site inference remains closed until identity verification and review are connected.

Rendered live in your browser by kami-engine

01 / INVESTIGATE

Review the code

Inspect code you own or are authorized to review for authorization gaps and input validation issues.

02 / VALIDATE

Ground the finding

Work through reachability, impact and false positives to produce evidence-backed findings.

03 / REMEDIATE

Close the loop

Prepare fixes, regression tests and reports for responsible disclosure.

A model built for investigation

TEXT / VISION / MoE

Qwen3.8 Flash Next

A real response has been verified on the dedicated deployment. Scaling up from idle can take several minutes. Access remains gated while identity verification and review are connected.

qwen3.8-flash-next-whitehacker

Model card

One verified identity. A free research workspace.

  1. Account

    Sign in with your existing account. A connected wallet alone is not verified identity.

  2. eKYC

    Identity verification runs on Stripe Identity only. Completing the document and selfie check at verify.stripe.com approves the account automatically (ekyc 365d, screening 24h, trust 60 points).

  3. AML / CTF

    Sanctions and PEP screening, risk review and additional checks where needed. Pending, expired or suspended records cannot infer.

  4. Free research

    Confirm authorization and research scope, then use the free allowance. Ongoing review keeps eligibility current.

$0 for verified researchers

Free allowance: 1,000 requests per day, up to 32,768 output tokens per request. Usage stops at the limit — no automatic charges. Past the allowance, monthly plans from $20/mo or prepaid AI credits keep requests running (never charged twice): top up any whole-dollar amount from $10 to $1,000 — one-time, 12-month validity, and larger packs carry a bonus ($50→51 credits +2%, $100→103 +3%, $500→525 +5%, $1,000→1,100 +10%) — or subscribe to a plan, or turn on automatic recharge.

Top up AI credits from $10

Before you spend a dollar, ground your work in our free open-data mirror — NVD CVE history (395k+), CISA KEV, EPSS, CWE, MITRE ATT&CK and ad-network intelligence as Parquet, no account needed: knowledge.mithril.fund

Preparing for launch — the selected model returned a real response on the dedicated deployment. Site inference remains closed until identity verification and review are connected.

Research terms & data notice

Do not enter identity documents, face images, private keys or API keys here. Before provider setup, application data is not sent or saved.

Research workspace

New models, pricing and free-allowance changes, by email

Your address is recorded once in Mithril's sales system and used for these updates. No other use.

Research terms & data notice

For authorized code review, vulnerability triage and remediation. No unauthorized intrusion, credential theft, destructive activity or third-party research without permission. Identity verification does not grant research authorization.

First-party browser identity verification is in development. Documents, face images, names and birth dates will never be sent to the inference model. The service is designed to retain minimal verification status, expiry, evidence references, review and usage records. Verification intake remains closed until providers, retention periods, transfer destinations, deletion and appeal contacts are finalized and published.

PEP status or a possible name match is not automatically a finding of wrongdoing. Additional checks and human review are planned. Legal AML/CTF obligations depend on the operator and service jurisdictions.

Confirm Stable Principal

Sign in make you confirm Stable Principal. Dat na di first step wey you fit finish. Hosted apply no dey offered.

You fit finish am now: sign in, confirm di Stable Principal, then connect am to Kotobase.

Start with Base Account

Three places wey work dey run. Their boundaries still dey.

Di services connect without become one giant trust domain. Each authority still dey separately governed.

CONTROL + IDENTITY

Mithril

Passkey dey verify Stable Principal. Di control plane dey publish di topology and authority floor wey CLI deploy dey use.

auth.mithril.fund · api.mithril.fund

One release CID wey you fit run from different providers

One release CID dey fix di namespace head, definitions, raw Wasm, compile receipts, and reproducibility evidence inside one IPLD graph. Names and GitHub still remain discovery and provenance.

Bundle

Put definitions, Wasm artifacts and compile receipts together under one release CID.

Replicate

Store di same complete closure for no less than two independent storage origins.

Verify + Run

Check every byte and routed peer IDs, then use di release CID and export take run am.

Shell
# install and run the live Ed25519 + ML-DSA-65 reference package
kotoba package add kotoba-lang/reference-math@0.1.0 --catalog-cid bafkreidcy5stqvnyfpmud6ozz5qz3supd3r3uzk7glmntuv36ezliaxstm
kotoba package run kotoba-lang/reference-math  # 42

kotoba library inspect <name|CID|#hash> --store .kotoba/codebase --namespace demo

# dry-run by default
kotoba library publish --store .kotoba/codebase --namespace demo --hosted

# replicate one exact release closure
kotoba library publish --store .kotoba/codebase --namespace demo --hosted --dry-run false \
  --pqc-seed-file <ml-dsa-seed> \
  --provider east=https://east.example --provider-token-file <east-token> \
  --provider west=https://west.example --provider-token-file <west-token>

# qualification and execution are release-CID addressed
kotoba library verify ipfs://<release-cid> --store .kotoba/codebase \
  --provider east=https://east.example --provider west=https://west.example
kotoba library run ipfs://<release-cid> --entry answer --store .kotoba/codebase \
  --provider east=https://east.example --provider west=https://west.example

# rotate or revoke the Principal-pinned ML-DSA key; both finish with Passkey
kotoba pq-key rotate --current-pqc-seed-file <current> \
  --next-pqc-seed-file <next> --expected-epoch 1
kotoba pq-key revoke --current-pqc-seed-file <current> --expected-epoch 2

Post-quantum signatures na mandatory, no be optional. Publication need both Passkey session and ML-DSA-65 signature wey pin to di Principal. External authenticators and distributed qualification still dey separately verified boundaries.

Open di library catalog and dependency graph

From AI-written code go admitted computation

Write

Agents and humans dey write freely for readable, data-oriented code.

Admit

Kotoba dey check types, effects, capabilities, resources, and target support.

Bind

Di host and provider dey bind Passkey identity and resource-scoped grant.

Run

Mithril Database keeps artifacts and receipts; Mithril AI and Itonami perform admitted work.

Shell
kotoba research --scope <approved-scope> --task code-review
# requests are admitted by Mithril before inference

Boundaries dey make connection possible.

To discover service no mean say you don give am authority

Storage, compute and agent work still get their own separate authority boundaries. Receipt dey record each origin separately. Existing Passkeys no dey move automatically; di new RP need Principal link wey dem don verify.