Is Toknbase an AI agent or a secrets manager?

Both — the vault is the engine, the agent is the product. Toknbase is an AI SecOps agent that actively scans repos, fixes exposed secrets, opens PRs, verifies every fix, and logs each action with a tamper-evident on-chain receipt. The vault stores secrets with zero-knowledge encryption so even the agent cannot read the values it protects.

What can the AI SecOps agent do?

The agent scans public and private GitHub repos for exposed credentials, monitors continuously on every new commit, proposes fix PRs to replace hardcoded secrets with secure runtime calls, verifies fixes are gone after application, checks whether found credentials have appeared in known breach databases, and sends a weekly CISO briefing with your full security posture. Every action is logged on-chain with a cryptographic receipt.

Can Toknbase — or its employees — read my secrets?

No. Client-side encryption means we store only ciphertext. We never have your encryption key. Even a Toknbase engineer with direct canister access would only see encrypted bytes. This is a cryptographic guarantee, not a policy promise.

How do I use Toknbase with AI agents like Claude or Cursor?

Toknbase provides a hosted MCP (Model Context Protocol) endpoint at https://4wj64-piaaa-aaaan-q5q7q-cai.icp0.io/mcp. Add it to your Claude Desktop, Cursor, or Windsurf config as a JSON block — no installation, no npm package, no local server required.

What is the pricing for Toknbase?

Toknbase offers three plans: Free ($0 — 1 seat, 50 secrets, 3 scans/month, free forever), Pro ($19.99/seat/month or $15.99/seat/month billed annually — unlimited seats, private repo scanning, AI auto-remediation, full MCP endpoint, pre-commit hook), and Enterprise ($499.99/month flat — everything in Pro plus SSO/SAML, custom detectors, dedicated support, DPA, and 99.9% SLA). Annual billing saves 20% on Pro.

Can I change plans at any time?

Yes — upgrade, downgrade, or cancel at any time. Upgrades take effect immediately. Downgrades apply at the next billing cycle.

Is there a free trial?

The Free plan is free forever and includes public repo scanning and an on-chain audit trail — no credit card required. Pro and Enterprise plans can be started immediately with a paid subscription.

How does Toknbase compare to Doppler or HashiCorp Vault?

Three key differences: (1) It's an AI agent, not a passive vault — it actively watches repos, finds leaks, and opens fix PRs without being asked. (2) Zero-knowledge encryption — Toknbase cryptographically cannot read your secrets, even under subpoena. (3) On-chain audit trail — every agent action has a tamper-evident cryptographic receipt. No other secrets management tool provides all three.

How is Toknbase different from using .env files?

.env files store secrets on disk in plaintext and are frequently committed to version control by accident. Toknbase encrypts secrets in your browser before they leave your device, stores only ciphertext, actively scans your repos for accidentally exposed values, and can automatically open a fix PR when it finds one. It's a proactive agent, not a static file.

Can Toknbase scan private GitHub repos?

Yes. Public repos can be scanned instantly with no signup. To scan private repos, connect your GitHub account via OAuth — Toknbase requests the minimum required permissions and uses a short-lived access token. Private repo scanning is included in the Pro plan.

What happens if a secret is exposed in an old commit?

Removing a secret from a file does not remove it from git history — anyone with access to the repository can still find it in old commits. Toknbase scans git history for exposed secrets, not just the current state of the code. When it finds one, it flags it as a historical exposure and recommends rotating the credential immediately, since it must be treated as compromised.

Does Toknbase scan for code vulnerabilities?

No. Toknbase specializes in credential exposure detection — API keys, tokens, and secrets hardcoded in your code. For code-level vulnerability scanning, Snyk and Semgrep are the right tools. Most teams use both.

Can Toknbase see my secret values?

No. Secrets are encrypted before leaving your browser using AES-256-GCM. The scanner detects credential patterns but masks values immediately — Toknbase never stores or transmits plaintext secret values.

Are scan results stored on-chain?

Scan metadata and finding summaries are logged on the Internet Computer with tamper-evident hashing. Secret values are never stored in the audit log.

New to Toknbase?Get set up in 5 minutes →

SCOPE & SECURITY · REFERENCE

What Toknbase scans for

Credential exposure detection — not code vulnerability scanning. Toknbase finds exposed API keys, tokens, and secrets hardcoded in your repos.

What Toknbase finds

Toknbase scans source code, config files, CI/CD workflows, and git history for exposed credentials that should never be in your codebase.

API keys and tokens — AWS, Stripe, GitHub, OpenAI, Google Cloud, Twilio, and 60+ more patterns
Hardcoded credentials in source code (.py, .js, .ts, .go, .rb, etc.)
Secrets in .env files committed to repos (even if later deleted from the latest commit)
Tokens in CI/CD workflow configs (.github/workflows/, .gitlab-ci.yml, .circleci/config.yml)
Credentials in git history — every commit, every branch, including deleted files
Private keys and certificates (RSA, EC, PEM format)
Database connection strings with embedded passwords
Bearer tokens and OAuth client secrets in code
Hardcoded passwords in config files (Docker, Kubernetes, Helm charts)

What Toknbase does NOT scan

Toknbase does not scan for code vulnerabilities.

Logic bugs, CVEs, injection flaws (SQL injection, XSS, buffer overflows), and insecure code patterns are outside Toknbase's scope. For that,Snyk andSemgrep are the right tools. Most teams use both: Snyk/Semgrep for code security, Toknbase for credential exposure. They complement each other perfectly.

✓ Toknbase's lane

✓Exposed API keys and tokens
✓Hardcoded credentials
✓Secrets in git history
✓CI/CD workflow credential leaks
✓Private keys and certificates

✗ Not Toknbase's scope

✗SQL injection and XSS vulnerabilities
✗Dependency CVEs and outdated packages
✗Insecure coding patterns
✗Buffer overflows and memory issues
✗Code logic bugs

Use Snyk or Semgrep for these.

Zero-knowledge model

Secrets are encrypted before leaving your browser. Toknbase never sees plaintext values — this is an architectural constraint, not a policy.

Algorithm

AES-256-GCM

KDF

HKDF (v5)

Encryption

Client-side

Secrets are encrypted in your browser — no server-side key material, ever.
The scanner detects credential patterns in file content but masks values immediately.
Every scan result is stored with the value masked — never plaintext.
Decryption only happens locally, when you explicitly view a secret in the dashboard.

Compliance and verification

Every Toknbase action is logged with a tamper-evident hash chain on the Internet Computer — independently verifiable by anyone.

Every action (found, fixed, rotated) is logged on-chain with a cryptographic receipt.
The hash chain cannot be altered retroactively — tamper-evident by design.
Audit logs can be exported as CSV or PDF for compliance reviews.
Verification endpoint: any exported chain tip hash can be independently verified.
Secret values are never stored in the audit log — only metadata (action, timestamp, actor).

Note

Verify any audit export attoknbase.net/verify— paste any exported chain tip hash and get an instant verification result, no account needed.

What Toknbase cannot see

By design, Toknbase has zero visibility into your actual secret values. This is not a policy — it is an architectural constraint enforced by the encryption model.

What we store	What we cannot access
Encrypted ciphertext blobs	Plaintext secret values
Secret names (metadata)	Encryption keys or passphrases
Audit log entries (action, timestamp, actor)	Master passwords (there are none)
Your Internet Identity principal	The content of your secrets

AI agent exposure risks

AI coding assistants can read your development environment. Here is exactly when your secret values are and are not at risk.

Values NOT exposed

·Using Toknbase MCP server with a scoped agent token
·Secrets stored in Toknbase vault (not in .env files)
·Toknbase servers never see plaintext — encryption is client-side only
·Agent access is user-controlled: you choose which secrets MCP can reach

Risk scenarios

·Pasting a raw API key directly into a chat window
·Storing secrets in .env files the AI agent can read
·Using a full_access token with no folder restrictions
·Not reviewing audit logs for unexpected agent access

Agent token best practices

Follow these practices to minimize exposure surface when connecting AI agents to your Toknbase vault.

Scope	Can do	Cannot do
read_only	List secret names, list folders	Create, update, delete, access values
read_write	All read_only + create and update secrets	Delete secrets, folder management
full_access	All read_write + delete, batch ops, audit log, retrieve values	Exceed token's folder restrictions

Use read_only scope for AI agents that only need to reference what secrets exist.
Apply folder restrictions to limit agent access to only the secrets it needs.
Rotate agent tokens regularly — especially after ending a project.
Review the audit log weekly to spot unexpected access patterns.
Never share an agent token in a public repo, chat, or pull request comment.
Create separate tokens per AI tool — never reuse the same token across editors.

Authentication

Toknbase uses WebAuthn via the Internet Computer's Identity system. There are no passwords — authentication uses hardware security keys, biometrics (Face ID, Touch ID), or device passkeys.

No password database — there is nothing to breach.
WebAuthn is phishing-resistant by design — credentials are bound to the domain.
Your Internet Identity principal is derived from your device key, not a server.
ICP's identity infrastructure is decentralized and operated by independent node providers.

Audit log

Every action on a secret — creation, update, deletion, access attempt, rotation — is recorded in a cryptographic audit log stored on the Internet Computer canister. These records cannot be altered or deleted.

On-chain storage makes the log tamper-proof by design.
Every entry records: action type, actor principal, timestamp, and affected secret.
Export the audit log as CSV for compliance workflows.
Audit log entries are anonymized (not deleted) when an account is closed.

SIEM integration

Forward audit events to any SIEM, log aggregator, or security monitoring platform via a structured webhook.

Generic webhook works with Splunk HEC, Datadog Logs, Elastic Logstash, Sumo Logic.
Events are batched (up to 50 per flush) to avoid overwhelming receiving endpoints.
Category-based filtering: forward only secrets activity, access events, or threats.
All SIEM delivery activity is itself recorded in the audit log.

Start with zero-trust credential security today

All features included. No credit card required.

View Quickstart →