Uniqent ☉

Any brain, any agent. Package an AI agent's whole brain once — persona, MCP stack, skills, memory, config — into one open, signed .uniqent file, and install it into whatever framework you run (Claude Code, Hermes, OpenClaw) in one command.

Think n8n, but for whole agents: a visual builder to assemble the thing, and a portable artifact you can install anywhere. Uniqent is the builder + packager + translator + installer — it sits above the frameworks so one brain travels between all of them.

▶ Try it in one command

No clone, no config, no API key — install a complete research-analyst brain into your agent:

npx @uniqent/cli try research-analyst

It auto-detects your framework, installs the persona + skill + web-fetch MCP + a linked memory graph, then tells you exactly what to ask. Run npx @uniqent/cli try --list to see the featured brains.

---

What is Uniqent?

Today an agent's "brain" is locked inside whatever framework you built it in — there's no portable unit for "a whole agent." Uniqent makes that unit and the workflow around it: build → package → share → install. You compose a brain in Uniqent Studio (a local-first visual builder) — or capture one you already have — and export a single signed .uniqent bundle that carries no secrets. Anyone installs it into the framework they run; a per-framework adapter translates the brain into that framework's native layout and asks only for the recipient's own credentials.

What you can do

• 🚀 Try a brain instantly — npx @uniqent/cli try <name> installs a featured, signed, creds-free brain into your agent in one command.

• 🧩 Build one visually — compose persona, MCP stack, skills, memory, channels, and flows on an n8n-style canvas in Studio, with credentials wired as "needs" edges, then export one .uniqent.

  

  Memory isn't a flat list: write facts with Obsidian-style [[entities]] and #tags and Studio parses them into an interactive memory brain — a force-directed graph (zoom / pan / drag, plus a 3D mode) of how facts, people, and topics connect.

  

• 📥 Bring what you have — uniqent export captures a running agent; uniqent import-vault turns an Obsidian / "second-brain" vault into a brain (SOUL.md → persona, USER.md → profile, MEMORY.md + notes → memory, [[wikilinks]]/#tags preserved).

• 🔁 Install anywhere — the same bundle installs into Claude Code, Hermes, or OpenClaw; the adapter translates it and resolves your credentials locally.

---

How it works

Build → Pack → Share → Install — one open bundle, any framework:

1. Build a brain in Studio (or capture an existing agent / vault).
2. Pack + sign it into a .uniqent (Ed25519 signature over a content digest) — a fail-closed secret-scan guarantees no API keys travel in the bundle.
3. Share it as a raw file or URL (e.g. straight from GitHub) — no hosted service required.
4. Install it: the adapter verifies the signature, shows a permission sheet, resolves your credentials locally, dry-runs in a sandbox, then writes the framework's native layout.

What makes it defensible — four things .dotagents-style sharing lacks:

• Install is a translation, not a copy. One canonical format → per-adapter native output. When a target can't hold something (e.g. Hermes' memory limits) it truncates/transforms and reports exactly what changed. Lossy is acceptable; silent loss is not.
• Secrets never travel in a bundle. It carries the wiring (which MCP server, transport, auth type), never the keys — safe to post publicly and still instantly runnable.
• No hosted dependency. Install from a file or URL; a registry is optional convenience.
• Trust is first-class. Signing, a permission sheet, a redactable memory preview, and a sandboxed dry-run — all in v1.

A "brain" = everything that makes an agent that agent

Part	What it is
Persona	personality, voice, role, goals — the identity
About	a README + avatar describing the brain (travels in the bundle)
Stacks (MCP)	the MCP servers it can use (GitHub, filesystem, web, …) and which tools
Skills	reusable, cross-agent SKILL.md capabilities
Memory	durable facts, decisions, preferences, a user/agent profile
Tools	native built-ins it has on (web search, browser, code exec, …)
Automations	scheduled/triggered tasks (e.g. a daily briefing)
Channels	where it's reachable (Telegram, Discord, Slack, …)
Config	model/provider prefs, autonomy level, allowlists

Install into any agent

Framework	How the brain lands	Status
Claude Code	skills → .claude/skills, persona+memory → AGENTS.md, MCP → .mcp.json	✅ v1
Hermes	persona → SOUL.md, bounded memory (prioritized + trimmed, reported), MCP/channels → hermes.json	✅ v1
OpenClaw	persona → SOUL.md, memory → MEMORY.md, skills → skills/, MCP/channels → openclaw.json	✅ v1
Codex · Cursor · Gemini	—	planned

---

CLI

npm i -g @uniqent/cli      # or use npx @uniqent/cli <command>

Command	What it does
try <brain>	One-command install of a featured brain (auto-detects your framework). --list to browse.
install <file|url|slug>	Install a brain into --target <claude-code|hermes|openclaw> (--root <dir>, --cred ref=val).
inspect <file>	Show a bundle's manifest, components, and permissions.
export --root <dir>	Capture a running agent into a .uniqent (auto-detects the framework).
import-vault <dir>	Turn an Obsidian / second-brain vault into a signed brain.
pack <dir> · validate <dir|file>	Build / check a brain from a source directory.
search <q> · hub <mcp|skills> <q>	Find brains in a registry index; discover MCP servers + skills across hubs.
keygen · sign <file>	Generate an Ed25519 keypair; sign a bundle.

A few real flows:

# Install a brain into your framework, resolving your own credentials locally:
uniqent install my-brain.uniqent --target claude-code --root .

# Install straight from a raw URL (no registry needed):
uniqent install https://example.com/dev-powerpack.uniqent --target openclaw --root .

# Or by slug from any hosted index.json:
export UNIQENT_REGISTRY=https://raw.githubusercontent.com/RiggdAI/uniqent/main/registry/index.json
uniqent search coding && uniqent install dev-powerpack --target hermes --root . --cred github_pat=…

Studio (visual builder)

pnpm --filter @uniqent/studio start   # local-first; open the URL it prints

Build from source

Requires Node 22.13+ and pnpm (corepack enable provisions the pinned version).

pnpm install && pnpm build && pnpm test

---

Docs

Doc	What's in it
docs/SPEC.md	The .uniqent bundle-format reference (generated from the zod schema).
docs/BUILD_PLAN.md	Full engineering spec + milestone plan.
docs/UX-REVIEW.md	CLI UX review + roadmap of gaps.
docs/CONTRIBUTING.md · docs/GOVERNANCE.md	How to contribute + project governance.
docs/SECURITY.md	Security policy + disclosures.

Status — pre-1.0, core loop works today

• Spec · core · builder — schema, bundle read/write + validation + secret-scan + Ed25519 signing, and a framework-agnostic engine to assemble a brain. ✅
• Studio — local-first React canvas to build a brain and export a signed .uniqent. ✅
• Bring what you have — uniqent import-vault + uniqent export. ✅
• Install — three adapters (Claude Code, Hermes, OpenClaw) + the CLI + Studio's Install button; the same signed .uniqent installs into all three. ✅
• Distribute — @uniqent/cli on npm; example brains in examples/; a file-based registry; memory-pack publishing. ✅
• Next — a hosted registry (accounts/web search) + a uniqent:// web "Install" handoff; Codex/Cursor/Gemini adapters.

License

• Code (CLI + adapters + core): Apache-2.0 (LICENSE)
• Spec text + schema: CC0 (LICENSE-SPEC) — any framework can implement an adapter without asking permission.

Contributing & community

Contributions welcome — see docs/CONTRIBUTING.md and docs/GOVERNANCE.md. Report security issues via docs/SECURITY.md.