- Category
- Developer Tools
- Rank
- No. 1661Tools index
Previous survey · No. 1672 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- thedotmack
- GitHub
- 8 stars
- Date
About
For the times your AI needs to act humanish. A single-use, scoped credential broker for AI chats.
What it does
Humanish lets a person sign into a web service through a remote Chromium session, saves that browser session encrypted, then lets a Claude chat perform a predefined action with it. The user opens a short link, copies its instructions into the chat, and receives the action result there.
Why it's ranked here
The security boundaries are thoughtful: browser sessions stay behind the action runner, links expire after one visit, and each bearer permits one defined operation. However, the repository describes an early MVP, and the supplied files show plans and architecture rather than implementation or tests. It is promising design work, not yet strong evidence of a finished tool.
What's good
The design keeps long-lived browser credentials out of chat history. Saved sessions use authenticated encryption, while each action runs in a fresh browser context. Atomic link consumption avoids the consistency weakness of a basic key-value store. The first workflow is narrow and understandable: create a NotebookLM slide deck, publish it, and return the resulting URL.
Tradeoffs
The MVP supports one hardcoded action and one user per dedicated server. Browser sessions can expire, session storage is not captured, and reauthentication requires another interactive login. The flow also depends on the user opening a link and pasting markdown into Claude because direct fetching would strip the instructions. Operating it requires Cloudflare, Fly.io, a browser container, email authentication, and downstream services.
How to use it well
Humanish fits an operator who wants a chat to run a tightly defined task against a service authenticated through browser cookies. Use named profiles, short lifetimes, and the smallest possible action catalog. It does not replace a general automation platform, an API integration layer, or a multi-tenant credential service. Treat the current repository as an architecture to evaluate or implement, not a proven deployment.
Technical notes+
README.md describes a Cloudflare Worker broker backed by a Durable Object and KV, plus a Fly.io VM hosting neko Chromium and a Node action runner. plans/2026-05-14-humanish.md specifies TypeScript packages under packages/broker and packages/box, AES-GCM profile encryption with profile-name AAD, Playwright storageState capture including IndexedDB, loopback-only runner binding, and NOTEBOOKLM_AUTH_JSON for passing decrypted state without a disk write. reports/2026-05-14-humanish-server.md explains the human-copy flow and named browser profiles. The supplied repository text contains no implementation source, package manifests, deployment configuration, or tests, so these mechanisms are documented design rather than independently demonstrated behavior.
Observed
- License
- Apache License 2.0
- Planned primary language
- TypeScript for the Cloudflare broker and Node action runner
- Interfaces
- Web dashboard, browser-rendered skill links, and an action endpoint called by Claude
- Deployment surface
- Cloudflare Worker and Durable Objects at the public tier; a single Fly.io VM at the private tier
- Initial action surface
- NotebookLM slide-deck generation followed by publishing to wooow.now
- Installation surface
- No end-user shell or local setup is intended; no install command or published package is documented
- Repository structure
- The supplied repository text shows documentation and a license, but no implementation source or tests
Read from README.md, LICENSE, plans/2026-05-14-humanish.md, reports/2026-05-14-humanish-server.md.
What it can do
Generate single-use credentials for AI chat sessions
AI chat request → Temporary authentication credentials
Scope AI permissions for specific conversations
Permission requirements and conversation context → Limited-scope access credentials
Broker secure authentication between users and AI systems
User identity and AI service requirements → Authenticated connection
Expire credentials after chat session ends
Active credential and session status → Revoked/expired credential
Manage temporary access tokens for AI interactions
Access request and security parameters → Time-limited access token
Tags
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.