01 / PlannerQueuedRunningDone
Turns a sentence into a plan
Reads the request, splits it into ordered steps and decides which specialists the job actually needs.
- Breaks work into typed steps
- Wakes only the agents it needs
- Replans when a step fails
Specialists that work like a team.
Ordin wakes only the agents a job needs, hands each a scoped brief and keeps them in one shared memory, so the work moves like a well-run team.
Ordin
reconciles Q3 invoices
drafts vendor emails for sign-off
triages the support queue
answers security questionnaires
writes the weekly revenue report
chases overdue contracts
migrates a thousand tickets
checks every figure twice
books the quarterly review
cleans the CRM
summarises yesterday's calls
flags anything over budget
updates the roadmap in Jira
files the expense claims
One job, handed down the line.
Every specialist picks up exactly where the last one stopped, from the same shared memory.
02 / ResearcherQueuedRunningDone
Finds the facts first
Pulls records, documents and contracts through typed connectors, and cites every source it used.
- Searches your connected tools
- Cites every source
- Never writes, only reads
03 / AnalystQueuedRunningDone
Does the maths you can check
Reconciles, compares and flags exceptions, then shows its working line by line so anyone can audit it.
- Reconciles and compares
- Flags anything off-policy
- Shows its working
04 / CoderQueuedRunningDone
Writes code that gets tested
Writes the scripts a job needs, runs them in a sandbox and only ships what passes its own tests.
- Sandboxed execution
- Tests before it ships
- Diffs you can review
05 / CriticQueuedRunningDone
Checks the work before you do
Re-runs every figure against its source and sends anything that does not match back for a second pass.
- Re-checks every figure
- Catches drift and hallucination
- Blocks unverified output
06 / CourierQueuedRunningDone
Delivers, with permission
Drafts the emails, tickets and updates, and queues anything outbound for a person to approve.
- Drafts in your voice
- Queues outbound for sign-off
- Posts where your team works
Watch them talk it through.
Every handoff is a message you can read, from the plan to the last draft waiting for your sign-off.
Run 8821 / #finance-ops5 agents
Prompt
Waiting for a job...
Coordination, handled.
Handoffs, memory, models and budgets are the runtime's job, so your agents can stay focused on theirs.
- 01
Shared memory
Every agent on a job reads from the same scoped context, so nothing gets lost in a handoff.
- 02
Model per agent
Give the planner a large model and the courier a fast one. Swap providers without rewriting.
- 03
Parallel by default
Independent steps run at the same time, so a nine-step job finishes in seconds, not minutes.
- 04
Scoped by design
Each agent only sees the tools and data its brief names, and nothing else.
- 05
Bring your own
Wrap an existing script or model as a specialist and it joins the team like any other.
- 06
Every step traced
See which agent did what, with which inputs, at what cost, and replay any step.
Define your own specialist.
A brief, a model, the tools it may touch and the policies it must follow. That is a whole agent, and it joins the team like any other.
- Any model with an API
- Scoped tools and memory
- Policies travel with it
auditor.tsTypeScript
import { agent } from "@ordin/sdk"; export const auditor = agent({ name: "auditor", brief: "Check vendor invoices against contracts", model: "large", tools: ["ledger.read", "contracts.search"], memory: { scope: "customer" }, policies: ["read-only", "spend-ceiling"], budget: { usd: 0.02 },});Agent auditor registered3 tools / 2 policies / USD 0.02 cap
Hand it the job. Keep the rules.
Free for the first 10,000 runs. Bring your own model, keep your own data, ship your first agent this afternoon.