I · Mission
Define a mission
Natural-language objectives become scoped, traceable execution plans — with explicit dependencies, not an opaque prompt.
Mandate · Issue Nº 2 · MMXXVI
The AI software factory — installed, not sold.
Authority over every agent, every action — granted in writing.
Mandate is a governed AI work control plane, delivered as your engineering stage. Missions, agents, policy and approvals — coordinated from one surface, with a signed record of every delivery. Your team keeps the product, the customers and the direction. The factory ships the roadmap.
Running in production
Mandate builds and ships Moklabs' entire portfolio today. We are our own first client.
Policy engine
Cedar — the same authorization engine AWS uses — decides every action.
Access
Private embedded-partner cohort. We install one company at a time.
I · The Problem
Your product is validated.
Your roadmap is waiting.
There is a class of company that grows faster than it can build: proven product, strong distribution — and an engineering bench that cannot ship fast enough. Hiring senior talent is scarce and slow; copilots only accelerate the already-strong. And putting agents to real work fails on trust: who verified it, what did it cost, who approved it? Mandate answers with a governed execution layer — installed as your factory.
II · The Solution
One factory, every
agent under mandate.
Every agent gets a mission, a scope, a policy and an approval path before it touches anything. You see the whole production line — what is running, what is waiting, what it costs — in one place. You hold the gates; the factory holds the pace.
codex.releasegithub.draft · branch isolated
Implementation
executing
$0.41
claude.reviewapproval gate — merge to main?
Code review
waiting
$0.08
claude.docsblocked on approval above
Changelog
blocked
$0.00
opencode.syncscope: linear.write · TTL 71h
Linear sync
idle
$0.13
4 agents · 1 mission · prod-eupolicy: cedar://moklabs/release · 0 actions out of scope
III · How it works
Mission → Mandate → Agents
→ Approvals → Audit.
It starts with one instruction. Mandate turns it into a scoped plan, grants the agents only what they need, and holds the risky steps for a human — speed inside the guardrails compliance actually requires.
Mandate will grant
- Agentscodex.release · claude.review · claude.docs
- Scopegithub.draft · linear.write
- Hold for humanmerge to main · publish changelog
- Expires72h · auto-revoke on completion
II · Mandate
Grant a mandate
Each run receives explicit permissions, constraints, a TTL and the approvals it requires before any action lands.
III · Agents
Execute through agents
Claude, Codex, OpenCode, custom agents and humans work in coordinated, isolated lanes — never stepping on each other.
IV · Approval
Pause at risk
Sensitive actions trigger approval gates. Low-risk work continues automatically. No false approvals, no rubber-stamping.
V · Audit
Audit everything
Every action, decision, cost and outcome is recorded in a single run timeline — built for operations and compliance.
IV · Under the hood
Six primitives every run
is built from.
Not features bolted on — the runtime is composed of these. Each one is addressable, observable, and enforced by policy end to end.
- I · MissionMission lifecycle
- Define objectives in natural language. Mandate decomposes them into traceable execution plans with explicit dependencies.
- II · RuntimeTask DAG execution
- Complex work as deterministic DAGs: parallel lanes, checkpoints, automatic reordering — never an opaque queue.
- III · ApprovalApproval gates
- Risky actions pause at human checkpoints. Low-risk work flows on its own. No false approvals, no rubber-stamping.
- IV · AgentsAgent runtime
- Specialized agents in isolated workspaces. Clear ownership, traceable side-effects, scope enforced by policy.
- V · ContextConversation intelligence
- Multi-turn context preserved across channels. Follow-ups stay coherent, auditable, and inside the writ.
- VI · AuditFull observability
- Decisions, runs, costs and outcomes on a single timeline — designed for operations and compliance, not dev curiosity.
V · The Writ
The permission document for autonomous work.
A writ defines who or what can act, under which scope, for how long, with which tools, and who approved it. It is the agent's legal document — nothing flows without it, nothing is lost from it.
Underneath: a Cedar policy engine (the same one AWS uses), an HMAC event log, per-run cost accounting, and an immutable audit trail.
Writ · MND-2026-04812
Quarterly review automation
Granted under cedar://acme/governance/quarterly · TTL 72h
Acme Corp
Tenant · prod-eu
Tenant · prod-eu
Granted to
codex.research, claude.legal, opencode.sync
Scope
board.review · linear.write · github.draft
Constraints
no-delete · no-merge · approval ≥ L4
Approved by
3 of 3 · directors signed
Actions
142 recorded · 03 pending · 0 out of scope
Sealed · 14:08 UTC
Issued 2026-04-29 · MMXXVI
VI · Decision record
When it's done, you don't
take our word for it.
Every mission closes with a signed record — what changed, who decided, what it cost. Replayable, exportable, admissible in a board review. The proof a founder needs from a factory they do not operate.
Decision record · MND-2026-04812 · sealed
Cut the 4.3 release & draft the changelog
sha256 · 9f2a…c41e
Requested
Ship 4.3 and publish the changelog by Friday.
Delivered
PR #812 merged · changelog live · 0 actions outside scope.
- 14:02Mandate granted to 4 agents · scope github.draft, linear.write
- 14:06codex.release opened PR #812 · +318 −44
- 14:11@matheus approved merge to main · gate cleared
- 14:13claude.docs published changelog · mission closed
Ratified by @matheus · 14:11 UTC$0.62 · 23 actions · 11 min end to end
VII · Use cases
Built for the companies whose
roadmap cannot wait.
I
Founders & operators
- The factory takes the line: implementation, review and delivery, under mandate.
- Approve at the gates that matter — read the certificate, not the diff.
- Per-mission cost, visible before and after every run.
- No senior squad to hire, no platform to learn.
II
The team you already have
- Your developers govern gates, review plans and set priority — not the conveyor belt.
- Every mission leaves a worked example: plan, diff, evidence.
- The team levels up by operating the factory, not by drowning in the queue.
III
Teams already running agents
- Bring your own agents under policy, approvals and audit.
- Enforce policy across models, tools and environments.
- Standardize observability, permissions and audit trails.
VIII · Position
Not another IDE.
The governance layer above your tools.
Mandate does not replace Cursor, Claude Code, Codex, Copilot, Linear, GitHub or Slack. It governs the autonomous work happening across them — whether we install the factory for you, or you bring your own agents under mandate. Not an agency, not a license: an engineering stage, in writing.
| Capability | AI coding tools | No-code automation | Internal scripts | [ Mandate. ] |
|---|---|---|---|---|
| Mission-based delegation | prompt only | fixed triggers | you write it | Plans, scoped & traceable |
| Multi-agent orchestration | single agent | no agents | you write it | Codex · Claude · humans |
| Scoped permissions | all-or-nothing | per-connector | your tokens | Per-run, per-tool, TTL |
| Human approval gates | — | manual step | you write it | Policy-triggered, blocking |
| Isolated execution lanes | shared session | — | — | Sandbox per agent |
| Audit-ready run history | chat scrollback | run logs | — | Signed timeline + diff |
| Per-run cost accounting | — | — | — | Per agent, per action |
| Policy enforcement | — | — | you write it | Cedar, at every action |
| Cross-tool execution | in-editor | if integrated | you write it | Linear · GitHub · Slack |
IX · Embedded Partner Program
A factory you can
account for.
We install Mandate inside a small cohort of companies with validated products and stalled roadmaps — and run the engineering stage with them, in the open.
You get
- A governed AI factory installed as your engineering stage
- Delivery, review and deploy under mandate from week one
- A signed certificate for every mission — evidence, cost, approvals
We ask
- A validated product with a real backlog
- A founder who reads the record and holds the gates
- A 30-minute fit call before we start
Work email only · We reply within two business days · No newsletter
Mandate.