The connector is live.Now talk to the OS.

01 Connected

The host chat is the interface. The OS is the organization.

After you add the Agentik-OS connector, stay in the host you already pay. Ask it to use Agentik-OS. The tools appear when the account has a paid monthly plan or an official OS unlock. A lifetime unlock does not create a cloud instance. OAuth can finish without a plan. Unpaid accounts then get no tools and no host jobs.

You can speak in plain language, or paste a tool call. Both reach the same tools. Agentik does not pay Claude, ChatGPT, Cursor, Codex, or Hermes. The host pays tokens. Agentik writes the job, the objects, and the gate. The host runs the model after you approve.

Start with talk. One call. The host answers from the pack. Do not list every OS, every agent, and every memory first.

Talk to growth-os about what we should do first.
Create a growth-os project for Client 1.
Remember that Client 1 refuses paid social.
Ask audience_researcher on growth-os who we should talk to first.

02 Talk

One call. A pack. A reply. Not a search.

talk is the conversation surface for every official OS, including OS that do not exist yet. It resolves the project, recalls only the fragments that match the message, and returns a compact Director pack. The host answers from that pack. Agentik still does not buy the tokens.

Do not call list_os, list_agents, list_instances, open_instance, or recall unless the pack names a next tool.

talk({ osId: "content-os", message: "Capture this idea as an object: [your idea]. Run the editorial line and stop before publish." })

talk({ osId: "growth-os", message: "Open one mission from product facts I already have: [facts]. Do not mobilize all 48. Stop before outreach." })

talk({ osId: "growth-os", project: "client-1", message: "Open one mission from product facts I already have: [facts]. Do not mobilize all 48. Stop before outreach." })

03 Projects

One OS. Many clients. Isolated stores.

Growth OS for Client 1 is not Growth OS for Client 2. Memory, objects, and app connections stay on that project. A $19 seat can hold five projects per OS. $99 can hold fifteen. Unlimited can hold forty.

list_projects({ osId: "growth-os" })

create_project({ osId: "growth-os", name: "Client 1" })

use_project({ osId: "growth-os", project: "client-1" })

04 Tools

talk is the conversation. The original six stay for formal jobs.

A workflow still starts through the Director. A team is a named agent. Instance tools reopen objects, memory, connections, overlays, resets, export, and hosted wake. Copy a call, then replace the brackets with your facts.

talk

Conversation pack. One call. No store tour

list_projects

Named projects for one OS. Isolated stores

create_project

Open a new client or mission on that OS

use_project

Switch the active project

list_os

Official OS this workspace may see

list_agents

Compiled roster. Only when you need a teamId

run_os

Formal host job envelope. Optional instanceId

ask_director

Formal Director job. Blocked until approve

ask_team

Formal specialist job. teamId from list_agents

approve

Human-gate receipt. Reset and wake wait here

list_instances

Cloud OS instances. Each row is a project

open_instance

Open one instance by id

list_objects

Page instance objects. Never the whole store

get_object

Read one object payload

put_object

Write one object. Kind from instanceContract

remember

Store a memory fragment on the project

recall

Page memory. talk already recalls what matters

forget

Delete one fragment. No ghost copy

list_connections

App connections. Toolkit plus ca_ id only

connect_app

Start an app connection. Returns an OAuth URL

overlay_patch

Private overlay. cannotOverride stays in force

reset_os

Blocked reset. memory, structure, or full

export_instance

Export, then optional real wipe

wake_os

$199 hosted Hermes request. approve required

talk --os growth-os "..."
talk --os growth-os --project client-1 "..."

list_instances()

open_instance({ instanceId: "<instance-id>" })

list_objects({ instanceId: "<instance-id>" })

get_object({ instanceId: "<instance-id>", objectId: "<object-id>" })

put_object({ instanceId: "<instance-id>", kind: "memo", title: "Reopen this memo", payloadJson: "{\"text\":\"facts I already have\"}" })

remember({ instanceId: "<instance-id>", scope: "work", body: "A fact that should survive the next session." })

recall({ instanceId: "<instance-id>" })

forget({ instanceId: "<instance-id>", fragmentId: "<fragment-id>" })

list_connections({ instanceId: "<instance-id>" })

connect_app({ instanceId: "<instance-id>", toolkit: "gmail" })

overlay_patch({ instanceId: "<instance-id>", patchJson: "{\"label\":\"private overlay\"}" })

reset_os({ instanceId: "<instance-id>", mode: "memory" })

export_instance({ instanceId: "<instance-id>", wipe: false })

wake_os({ instanceId: "<instance-id>" })

05 Operating systems

Ask which OS you can use. Do not invent an id.

list_os returns official packages this workspace may see, split into entitled and locked. Decision, Content, Growth, and Librarian are in the catalog. Studio is how you create or improve an OS on a paid monthly plan. A listing cannot grant an entitlement. You cannot talk to an OS that is locked.

1. content-os

   Content OS

   Director first, then Researcher as researcher.

2. growth-os

   Growth OS

   Director first, then Audience researcher as audience_researcher.

list_os()

npx @agentikos/os login
npx @agentikos/os install --host cursor
npx @agentikos/os os list
npx @agentikos/os call list_os
npm run mcp -- list-os
npm run mcp -- list-agents growth-os

curl -s -X POST https://mcp.agentik-os.com/api/mcp -H "Authorization: Bearer $AGENTIK_API_KEY" -H "Content-Type: application/json" -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_os","arguments":{}}}'

06 Director

Every conversation starts with the Director.

The Director holds the mission. It decides which teams move, which skills load, and when the work should stop. talk is that conversation. Do not skip it to spray prompts at every agent. ask_director is the formal job. It stays blocked until approve.

ask_director({ osId: "content-os", prompt: "Capture this idea as an object: [your idea]. Run the editorial line and stop before publish." })

ask_director({ osId: "growth-os", prompt: "Open one mission from product facts I already have: [facts]. Do not mobilize all 48. Stop before outreach." })

07 Agents

Call a named specialist. teamId is the agent id.

list_agents returns the compiled roster for one OS. Same ids as the public landing. Each agent declares an intelligence need (code, write, review, and the rest). The host runs the mapped model. Updating the catalog updates every OS that tracks latest. ask_team needs osId, teamId, and a prompt. teamId is an agent id, or a team id, from that roster. Unknown ids fail. Call list_agents instead of guessing.

list_agents({ osId: "content-os" })

ask_team({ osId: "content-os", teamId: "researcher", prompt: "Collect inspectable sources for this idea. Leave missing facts missing." })

list_agents({ osId: "growth-os" })

ask_team({ osId: "growth-os", teamId: "audience_researcher", prompt: "From these product facts, who first? Do not invent a customer." })

08 Gates

Formal work does not run until you approve.

talk answers from the pack. It does not open a job.run_os, ask_director, and ask_team open a job envelope. Each returns a jobId. reset_os and wake_os also wait on approve. Until that call, host LLM jobs stay NOT_RUN. Agentik does not silently call a cloud model. RELEASE stays BLOCKED.

1. 01

   Create the job

   Director, team, or run_os. You get a blocked envelope, not a finished answer.

2. 02

   Read the receipt

   Check the OS, the tool, and the instructions. Send it back if the brief is wrong.

3. 03

   Approve or reject

   approve with the jobId. The host then runs the model. Agentik never buys those tokens.

run_os({ osId: "growth-os", intent: "Design the next experiment from facts I already have." })

approve({ jobId: "<job-id>", approved: true })

09 Logic

An OS is a composition, not a pile of chats.

The Director holds the mission. A NanoTeam holds a specialty. Agents do the work. Skills are the methods those agents know. Workflows are the repeatable path. Data is the objects the OS keeps. Memory is what should persist after the run. Permissions say who may act. Evaluations say whether the work was good enough to keep.

One OS is useful. A stack becomes an organization. Decision can inform Content. Content can feed Growth. Growth can update a later OS. That cycle is the shape of the catalog, not a claim that every package is live. Studio is where you create a new OS on a paid monthly plan. A lifetime unit unlock cannot create OS.

Talk to the system. Do not rebuild the organization in every window.