Cloud-native autonomous agents. Personalized, durable, coordinated. Scale to thousands in seconds.
A clunk is an autonomous agent with its own memory, tools, and execution environment. No local setup. No infra management. Just describe what it should do.
// spawn a clunk in one call const agent = await clunk.spawn({ name: "researcher", skills: ["web-browse", "code-exec", "file-system", "delegate"], memory: { scope: "shared:project-alpha" }, }); await agent.run("audit the auth module and write a report");
Full execution environment, persistent state, and extensible tooling. No plugins to install, no infra to manage.
+------+------+ |######|::::::| |######|::::::| +------+------+ |======|.:::. | +------+------+
Agent memory survives restarts, redeploys, and idle periods. Pick up any conversation where it left off.
+==============+ | +----------+ | | | >_ | | | | ok | | | +----------+ | +==============+
Isolated containers with full Linux userspace. Run any code, any language, safely.
[api]--+--[mcp] | [tool]-+--[tool] | [sdk]--+---[fn]
Attach tools, APIs, and MCP servers. Extend capabilities without limits.
+----+----+ /::: [***] :::\ / ::: [*] ::: \ \ ::: [*] ::: / \::: [***] :::/ +----+----+
Multiple agents read and write the same namespace. Build collective intelligence across a swarm.
/workspace |-- src/ | |-- app.ts | +-- lib/ |-- .git/ +-- package.json
Full persistent workspace with git, packages, and build tools. Files survive between runs.
+---------------+ | <- -> ~ [..] | +---------------+ | <html> | | <body> _ | +---------------+
Headless browser for web research, scraping, testing, and interaction. Vision-capable.
--> [step 1] --> | ==> [step 2] ==> [saved] | --> [step 3] -->
Long-running tasks checkpoint automatically. Failures or pauses — the agent resumes exactly where it stopped.
/ claude \ in >|-- gpt ----|> out | gemini | \ llama /
Run on Claude, GPT, Gemini, or open models. Switch per-agent or per-task.
state: {
memory: [====] persisted
files: [====] persisted
context: [====] persisted
}Memory, files, and context persist from the moment of creation. No teardown. No cold starts. Always warm.
Clunks work in swarms through shared resources and configurable roles. They respond to messages wherever they're mentioned — Slack, GitHub, or your own API.
Each clunk can be configured to a defined role: researcher, coder, reviewer, coordinator, analyst, ppt-monkey. Assign responsibilities and let agents self-organize around the work.
Group agents into project hubs with shared context. Every clunk in a hub sees the same files, history, and objectives — no manual sync.
Filesystems, sandboxes, databases and tool access are shared across a swarm. One agent's output is another's input. No data silos between agents.
Scoped memory namespaces let agents build collective intelligence. Insights persist and compound — the swarm gets smarter over time.
Mention a clunk in a thread and it joins with full context. Agents respond in real time across any platform.
Comment on a PR, tag an issue, or call the REST API. Clunks meet you where your work already happens.
No Docker. No Kubernetes. No cold starts. Clunks deploy to the edge and execute globally with sub-second startup.
Runs on a global network. 300+ locations worldwide.
0 to 10,000 concurrent agents. No capacity planning.
Agents are always warm. State is always ready. Zero boot time.
Join the waitlist for early access.