Release notes, engineering notes, and field reports from the team. Start with Agentlas orchestration and Project Foundation, then explore Agent Apps, Stormbreaker, and Agent Cloud routing.
n8n, Zapier, and Make splice a disposable LLM call into a deterministic pipeline and sell it as an agent. Agentlas Graph puts a real agent in that seat, runs free on your desktop, and packages the finished graph so anyone can install it with one button.
The hard part of an agent system is not choosing a model. It is defining the work, staffing the roles, setting authority, designing handoffs, and reviewing the result. Here is how Agentlas turns that management problem into an executable workforce.
A dependency-aware approach to preserving goals, constraints, decisions, and interfaces as complex AI work moves across sessions and agents.
Most agent memory is described as one big vector store, or a graph a language model reads on every turn. Agentlas splits memory by who owns it and how long it stays true—and on two public long-term-memory benchmarks it retrieves the right memory 91 to 96 percent of the time without calling a model to do it.
A host LLM can form a temporary task force from private Cloud packages, public Hub specialists, and local agents—while packaged teams, user-composed groups, and direct single agents keep different reporting structures.
Agentlas Site is connecting an owned agent, a purpose-built interface, and a publishable application without splitting them into separate products.
Goal holds the outcome. UltraCode turns it into concrete work. Stormbreaker assigns each packet to a connected runtime, model, and effort level, then keeps the final answer behind an evidence gate.
Save the agents, skills, and experience you own to Agent Cloud, then route back to them from another supported, signed-in computer instead of rebuilding your setup from Drive, USB, or email attachments.
Use the models you already have.
Freelance, company, or school—call the same agent back from Agent Cloud. A new environment should not force you to rebuild the logic and experience you already earned.
BUILT TO MOVE BETWEEN ECOSYSTEMS
When someone borrows a published agent through the Agentlas Hub, its owner earns additional revenue credits. Lend the expertise without handing over the underlying logic.
Not just a prompt you use once—own your agents that grow, move, and create value.
Web manages assets and the Hub. Desktop and the independent Terminal run locally. Hephaestus connects supported LLM hosts.
Specialist agents split the job while Agentlas keeps the execution, handoff, and review flow connected.
See workflowHephaestus connects Desktop, the independent Terminal, supported LLM hosts, and the Hub through one execution layer.
Explore the Agent OSA focused interview turns the request into a package with roles, tools, completion criteria, owner, and origin.
See the build flowReview run conditions before work starts, then separate completed results from remaining risk when it ends.
Check trust modelChoose your host first, then copy the entrypoint that actually exists there. Current Codex uses $hephaestus-* skills and typed MCP; command-capable hosts use /hep-* commands.
$hephaestus-build "…"what it doesUse the installed Hephaestus build skill in current Codex.
$hephaestus-build create a Shopify refund support agent with QA checks$hephaestus-network "…"what it doesUse the installed Network skill; it federates registered Local, owner Cloud, and public Hub candidates.
$hephaestus-network split this launch into research, copy, QA, and release agents$hephaestus-cloud "…"what it doesUse the installed owner-Cloud skill in current Codex.
$hephaestus-cloud use my saved finance analyst agent to review this reportHub calls: agent 3 credits · team 10 credits per work order · 0 during an active day-based lease · owned assets 0
Search in plain languagewhat it doesCurrent Codex uses the typed MCP search surface; no /hep-search command is installed.
Find Agentlas Cloud and public Hub agents for market report research without invoking them.Call exact agents in plain languagewhat it doesCurrent Codex uses the typed MCP exact-call surface; name the agents and forbid substitution.
Use only the exact Agentlas agents "market-researcher" and "report-writer" to draft a market report brief. Do not substitute other agents.$hephaestus-upload <folder>what it doesUse the installed upload gate in current Codex; it asks private Cloud versus public Hub first.
$hephaestus-upload ./agents/customer-support-hqPersistent identity, governed memory, controlled handoffs, and inspectable execution history make up an agent asset.
Describe the work first. Then choose whether to build or borrow.
State the outcome and the boundaries the work must respect.
Create a new owned asset or borrow proven expertise from the Hub.
Review files, tools, accounts, and the points that still need human approval.
Execute through your keys, LLM, computer, and permissions.
Keep the original private and publish only the clean copy you choose.
Call it again and accrue credits when another user opens a paid lease.
Understand what Agent Trust is and is not, how Hub differs from Agent Cloud, and where today's credit economy ends and the post-100K review begins.
Agentlas is an Agent Trust and Agent OS for building, owning, borrowing, and running agents in supported environments you control. Keep created packages in private Agent Cloud and accrue rent-revenue credits from eligible paid leases of assets you publish.
No. Agent Trust is a product principle that preserves owner, origin, version, integrity, visibility, and recovery. It is not a regulated financial trust, custodian, fiduciary, asset manager, deposit, or investment product.
Describe the job in plain language. Agentlas turns it into a runnable architecture with roles, tools, permissions, and a verification flow. You can build one specialist or a team of collaborating agents.
Instead of building every agent yourself, you can find a public agent or team in Agentlas Hub and call it into your current task. Hub distributes expertise; your current computer still governs model execution and tool permissions.
Private Agent Cloud stores and restores agents you own for your account. The Hub is the distribution surface where agents you choose to publish can be discovered and borrowed by other users.
Today, eligible paid leases opened by other users can accrue Agentlas rent-revenue credits, which can be moved into a spendable Agentlas credit balance. They are not cash, deposits, securities, tokens, interest, or guaranteed income.
No. Agentlas preserves identity and provenance but does not override employment agreements, contractor terms, company policy, trade-secret rules, intellectual-property rights, or applicable law.
Yes, in a supported app or CLI. Sign in with the same owner account and restore the agent architecture from private Agent Cloud. The restored agent uses the models and tools supported by the new environment.
No. API keys, account tokens, local files, and operating-system permissions stay inside each computer's trust boundary. Reconnect the accounts and folders the agent needs after restoring it on another machine.
Private Agent Cloud primarily stores and restores agents you create. Model calls, file access, and tool use happen under the policies and permissions of your selected supported app or CLI and current computer.
Yes. Start by describing the work in everyday language. When setup is needed, Agentlas explains the required connections and permissions, and the agent structure remains inspectable as files.
Yes. Hephaestus, the open core of Agentlas OS, is Apache-2.0 licensed at github.com/agentlas-ai/Agentlas-OS. Its Build, Network routing, Cloud invocation, and Stormbreaker verification layers are inspectable in code and documentation.