Goals

Goals turn Reeve from a reactive assistant into a proactive operator. A goal is a persistent, multi-phase objective that survives session resets and gateway restarts — letting agents plan, execute, and monitor work over hours or days.

Why Goals?

Without goals, agents respond to messages and go idle. With goals, agents pursue sustained objectives:

"Run our product launch campaign. Budget: $200 on ads, target 500 signups by Friday."

Reeve creates a goal with phases — research, build creatives, deploy campaign, monitor and optimize — and works through them autonomously, reporting progress along the way.

How Goals Work

Every goal has:

• Title and description — What you're trying to achieve
• Phases — Ordered steps that run sequentially (research → build → deploy → monitor)
• Budget — Token and dollar caps to prevent runaway spending
• Checkpoints — State snapshots that survive context compaction
• Metrics — Tracked KPIs like signups, CTR, or revenue

Phase Types

Each phase can run in one of three modes:

Goal Lifecycle

created → active → [phases running] → completed
              ↓
            paused → resumed
              ↓
            failed / cancelled

Goals move through statuses: active, paused, completed, failed, cancelled. Each phase within a goal tracks its own status independently.

Creating a Goal

Via conversation

Just tell your agent what you want to accomplish:

"I need to increase newsletter signups to 1,000 by end of month."

The agent creates a goal, breaks it into phases, and starts working.

Via CLI

reeve goal create

The interactive wizard walks you through title, phases, budget, and deadline.

Via agent tools

goals({
  action: "create",
  goal: {
    title: "Increase newsletter signups",
    description: "Get to 1,000 signups by end of month",
    phases: [
      { name: "Audit current funnel", type: "once" },
      { name: "Optimize landing page", type: "once" },
      { name: "Monitor & iterate", type: "loop" }
    ],
    budget: { maxCost: 50, approvalGate: 25 }
  }
})

Budget and Approval Gates

Every goal can have spending limits:

• Token budget — Maximum tokens the goal can consume
• Dollar budget — Maximum cost in dollars
• Session budget — Maximum number of agent sessions
• Approval gate — Pause the goal and ask for human approval at a spending threshold

When a goal hits its approval gate, it pauses automatically. You'll receive a notification asking whether to continue. Approving raises the gate by 50% and resumes the goal.

// Check remaining budget
goals({ action: "budget_check", goalId: "goal_abc" })

// Approve past the gate
goals({ action: "approve", goalId: "goal_abc" })

Monitoring Goals

CLI

reeve goal list              # All goals
reeve goal status <id>       # Detailed view with phases, budget, logs
reeve goal log <id>          # Recent activity log

Agent tools

// List all goals
goals({ action: "list" })

// Get goal details
goals({ action: "get", goalId: "goal_abc" })

// Add a log entry
goals({ action: "log", goalId: "goal_abc", message: "Landing page A/B test launched" })

// Update a metric
goals({ action: "metric", goalId: "goal_abc", name: "signups", value: 342 })

// Save a checkpoint
goals({ action: "checkpoint", goalId: "goal_abc", summary: "Phase 2 complete — creatives deployed" })

Service Loops

For intensive monitoring phases, goals can run service loops — fast execution cycles with intervals as short as 10 seconds:

// Start a monitoring loop
goals({ action: "start_loop", goalId: "goal_abc", intervalMs: 300000 }) // every 5 minutes

// Stop the loop
goals({ action: "stop_loop", goalId: "goal_abc" })

Service loops are ideal for campaign monitoring, real-time metric tracking, and automated optimization.

Example Use Cases