Agent Creator
Sasha does not make reports. Sasha creates agents, and the agents make the reports, recommend what to do and, with your approval, do it.
Most AI tools write a report when you ask. Agent Creator creates an agent that does the job for you. The agent goes out and collects the facts, writes the report, tells you what to do next and, with your approval, does it. It works on a schedule and it does not wait for you to ask.
Sasha already knows your business: your systems, your people, who decides what and how often you review each thing. Agent Creator uses that knowledge. Each agent is built for your business, not for a generic one.
What a Sasha agent is made of
Every Sasha agent has three parts. Each part has one job, and each can change without the others.
| Part | What it is | What it means for you |
|---|---|---|
| Its mind | The AI, the agent's instructions, its brief and its reading list | How it thinks is written down, so you can read it and change it. When AI models improve, the mind improves, and nothing else must be rebuilt. |
| Its hands | The tools an admin lets it use, for example read-only access to your cloud bill | It can only do what you approved. Your passwords and keys stay locked inside Sasha. The agent can use a tool, but it cannot read or copy the key. |
| Its notebook | A permanent record of everything it found, did and was told | It never forgets and never repeats itself. You can ask "why did you say this?" about any line, and it shows the record. |
Today each agent keeps a record of every run. The full notebook, with every item, every decision and every mark you make, is planned for all agents.
How it works
The example is the admin of Northwind Advisory, a small consultancy.
- The admin opens Apps and chooses an agent, for example the Finance Agent. They click Set it up.
- Sasha reads what it already knows about Northwind. It says what it can already see and what is probably missing: "A consultancy like yours usually spends on eleven things. I can see six of them."
- Sasha asks only about the gaps, a few questions at a time, and connects each source read-only.
- Sasha builds Northwind's own agent: its name, its owner, its schedule, its thresholds and its words. It runs it once and shows where each figure came from.
- The admin checks the first report and approves it. From then on, the agent runs by itself.
Three levels of work
| Level | What the agent does | Example |
|---|---|---|
| 1. It reports | Collects the facts, checks them and writes the report | "September cost: 4,210 GBP, by product." |
| 2. It recommends | Gives at most three things to do, each with an owner and a request you can paste into Sasha | "Card fees went up 40%. Tell Sasha: compare our card fees with last quarter." |
| 3. It acts, with your approval | Prepares the action and waits for a person to approve it | "Cancel the unused test server?" Approve or dismiss. |
An agent never acts as a person. It runs as itself, it uses only the tools an admin approved, and an action that it may not take waits for a person in Held items.
The agents
| Agent | What it looks after | Status |
|---|---|---|
| Finance Agent | What you spend, by product; budgets; failed and late payments | Available now |
| Marketing Agent | Which channels bring leads and sales, and what each one costs | Available now |
| PR Agent | What the news and the market say about you, your clients and your competitors | In development |
| Operations Agent | The daily health of your services: errors, sign-ups, costs and stuck jobs | In development |
| Project Agent | Your projects: the plan, what is late, what is at risk and what waits for a decision | Planned |
| Sales Outreach Agent | Who to contact next, and a first draft of what to say | Planned |
| HR Agent | People, access and the accounts to switch on and off | Planned |
Each new agent uses the same mind, hands and notebook design. So the list grows without new platform work.
You will also be able to describe an agent of your own in plain words. See How to make your own agent.
Every agent has a reading list
Status: planned. A good expert can tell you which books and methods taught them. A Sasha agent will do the same. When Agent Creator builds an agent, it will also make the agent's reading list: the methods, frameworks and sources the agent uses to think. A person can review and add to the list.
This will let you check that the agent's methods are current and ethical, and let the agent cite them when you ask why. How agents learn tells you more.
Why Sasha's agents are different
The mind
- It starts from your knowledge. Sasha already holds your documents, meetings and connected systems. The agent does not start from nothing.
- Scripts do the sums, not the AI. Tested scripts calculate the numbers and the scores. The AI explains them and judges only what a script cannot.
- You will see where the thinking comes from. Each agent will have a reading list that you can check.
The hands
- A person approves every action. An agent acts only within the authority an admin gives it.
- The agent cannot read your keys. Sasha keeps the passwords and keys. The agent uses the tool; the key stays with Sasha.
The notebook
- Every figure has a source. A figure that Sasha could not read shows as "not read", never as zero.
- It remembers. It does not report the same thing twice, and it measures change against what it saw before.