What happens to your agent when you cancel Claude?

Agentis gives your agents a permanent home. Memory, skills, identity, and work that survive any model swap. Self-hosted, open source, yours.

Then open your local workspace with agentis up.

See the system

Inside Agentis

Not a promise. A working organization.

Project 3gT7 runs on Agentis today. Explore its real workspace, agent fabric, Brain, and Agentic Apps—captured from the local product, not redrawn for marketing.

Project 3gT7 · live workspace

An organization you can see, question, and operate.

The workspace is the command surface for a real agent organization: an orchestrator, domain managers, specialist-owned apps, live status, and a direct line to the agent responsible.

  • Workspace-scoped authority
  • Live agent status
  • Apps attached to owners
1 / 4
01 / 04

The ownership model

Own the agent. Rent the model.

Models are powerful execution engines. They should not be the only place your agent’s identity, experience, and operating knowledge exist.

“A model can change overnight. Your agent should wake up as the same agent tomorrow.”
Orchyidentity · Brain · skills

Permanent agent

Execution runtime

Runtime changed. Agent continuity preserved.

The permanent Brain

A mind with structure—not an infinite transcript.

Agentis separates memory, grounded knowledge, learned procedure, and identity while keeping them on one durable substrate.

brain://orchy/memory

Experience that survives the session.

Candidate memories pass deterministic checks, formation judgment, and scope rules before they become durable. Useful experience compounds; noise does not silently become truth.

  • Formation pipeline
  • Private and shared scopes
  • Outcome-aware recall

Introducing Agentic Apps

The app is no longer a shell around the agent.The agent operates the app.

Agentic Apps package interface, data, staffed agents, orchestration, and goals into one durable product that can act, learn, and remain accountable.

Read the category definition
01

Typed data

Collections and schemas give agents durable state they can query and change safely.

02

Product surfaces

Interfaces expose the live product instead of returning another wall of generated text.

03

A staffed cast

Orchestrators, managers, and specialists remain accountable for the app they operate.

04

Operational logic

Workflows, goals, triggers, and evidence turn autonomy into inspectable execution.

Interface+Data+Agents+Workflows+Goals= Agentic App

The Workflow Engine

Completed is a runtime state. Accomplished is a verdict.

Agentis keeps execution status separate from objective truth. Every delivery can carry the trace, evidence, and acceptance result that justify it.

run.statusCompleted

All executable nodes settled.

Necessary. Not sufficient.
objective.verdictAccomplished

Acceptance criteria are supported by evidence.

Safe to deliver.
S

SWIFT / stage 1

Scope

Define the objective, evidence, limits, and acceptance contract.

When reality disagrees
  1. Retry safely
  2. Diagnose
  3. Replay the affected path
  4. Propose structural repair
  5. Request approval when boundaries change

One durable stack

Every harness. One Brain. Zero lock-in.

The Runtime Abstraction Layer normalizes what each harness can actually do. Routing is based on capability and policy—not a hard-coded vendor name.

Product layer
Agentic Apps Workflows Channels
Agent layer
Identity Brain Skills
Runtime Abstraction Layer
Affordance matching Model routing Policy
Claude CodeCodexCursorAntigravityHermesOpenClawHTTPLocal models
Your infrastructureSQLite · secrets · assets · agent homes · audit

Reach, with boundaries

Let agents act in the world without losing the operator.

Channels, integrations, MCP, A2A, webhooks, and email share the same connection, credential, policy, and audit layers.

ChannelsMCPA2AWebhooksEmailHTTP

Local-first by default

Runtime data stays under the Agentis data directory unless you deliberately connect an external system.

Bounded credentials

Encrypted connection secrets and explicit permissions keep reach separate from unrestricted access.

Approvals and budgets

Human gates, spend ceilings, and circuit breakers make autonomy observable and interruptible.

A durable audit trail

Runs, actors, tool use, costs, approvals, and evidence leave an inspectable operational record.

Run it on your machine

Give your agents somewhere permanent to live.

Node.js 20.10+ · macOS, Linux, or Windows PowerShell · no provider key required to boot.

thenagentis uphttp://127.0.0.1:3737

Questions, answered plainly

Before you install.

Agentis is usable today and still pre-1.0. APIs and product surfaces may continue to evolve as the ownership model hardens.

Is Agentis another model or chat client?

No. Agentis is the durable layer around the models and harnesses you already use. It owns agent identity, memory, skills, workflows, apps, approvals, and evidence while the execution runtime remains swappable.

Does Agentis send my workspace through Agentis servers?

No. The standard self-hosted runtime serves the dashboard, API, database, secrets, agent homes, and audit records from infrastructure you control. The model or harness you connect may have its own network behavior and policies.

Which systems can run underneath an Agentis agent?

The current adapter surface covers Claude Code, Codex, Cursor, Antigravity, Hermes, OpenClaw, HTTP agents, and OpenAI-compatible endpoints such as local model servers. Adapters normalize their capabilities behind one runtime contract.

What makes an Agentic App different from an agent workflow?

A workflow is execution logic. An Agentic App is the durable product around that logic: typed data, user-facing surfaces, a staffed cast, workflows, goals, permissions, and operational history.

What does “completed is not accomplished” mean?

A run can finish every node and still fail its real objective. Agentis records execution completion separately from evidence-backed accomplishment, so retries, repair, and delivery decisions are based on the result—not merely the absence of an exception.

What do I need to run Agentis?

Node.js 20.10 or newer on macOS, Linux, or Windows PowerShell. Install the CLI globally, run agentis up, then open the local URL printed by the server. No model-provider key is required just to boot the platform.