Token Budget Allocator

Overview

When you already know your total token budget and how important each section is, we split it across named items — honouring per-item minimums, optional maximums, and weights. Minimums are satisfied first; if they cannot all fit we return feasible=false with a shortfall instead of throwing. The remainder is shared by weight, capped at each item's max, with leftover whole tokens assigned by largest-remainder rounding (ties break in input order). Pure integer math — no tokenizer — do not use this to count tokens in text or to trim content.

Dividing a fixed context window by hand is error-prone. We apply one deterministic rule so your floors, caps, and weights stay consistent.

When to use it

Scope

Non-goals

• Does not count tokens in text
• Does not truncate or select content
• Does not apply hidden heuristics

Properties

Capabilities

Cost & billing

Flat rate per invocation

File handling

How we handle your data

Input schema

Output schema

API contract

curl -sS -X POST "$BASE/v1/execute" -H "Authorization: Bearer $KEY" -H 'Content-Type: application/json' -d '{"skillKey":"context.allocate_budget","input":{"totalBudget":128000,"reserve":8000,"items":[{"id":"system","min":500},{"id":"history","weight":2,"max":40000},{"id":"retrieval","weight":3}]}}'

import { Squillify } from "@squillify/sdk";

const sq = new Squillify({ apiKey: process.env.SQUILLIFY_API_KEY });

const { output } = await sq.execute("context.allocate_budget", {"items":[{"id":"a"},{"id":"b"},{"id":"c"}],"totalBudget":1000});

Errors

Other limits

Limitations

• Does not count tokens in text
• Does not truncate or select content
• Does not apply hidden heuristics
• Measuring how many tokens text actually needs
• Non-integer or probabilistic allocation

How to call

POST /v1/execute (requires authentication)

{
  "skillKey": "context.allocate_budget",
  "input": {
    "items": [
      {
        "id": "a"
      },
      {
        "id": "b"
      },
      {
        "id": "c"
      }
    ],
    "totalBudget": 1000
  }
}

Examples

{
  "items": [
    {
      "id": "a"
    },
    {
      "id": "b"
    },
    {
      "id": "c"
    }
  ],
  "totalBudget": 1000
}

{
  "warnings": [],
  "allocatedTotal": 1000,
  "distributable": 1000,
  "unallocated": 0,
  "skill": "context.allocate_budget",
  "feasible": true,
  "allocations": [
    {
      "id": "a",
      "allocated": 334
    },
    {
      "id": "b",
      "allocated": 333
    },
    {
      "id": "c",
      "allocated": 333
    }
  ],
  "shortfall": 0
}

{
  "totalBudget": 1000,
  "items": [
    {
      "min": 50,
      "id": "system"
    },
    {
      "id": "context",
      "weight": 3
    },
    {
      "id": "output",
      "weight": 1
    }
  ],
  "reserve": 100
}

{
  "warnings": [],
  "allocatedTotal": 900,
  "distributable": 900,
  "unallocated": 0,
  "skill": "context.allocate_budget",
  "feasible": true,
  "allocations": [
    {
      "id": "system",
      "allocated": 220
    },
    {
      "id": "context",
      "allocated": 510
    },
    {
      "id": "output",
      "allocated": 170
    }
  ],
  "shortfall": 0
}

Recommended use cases

• Divide a context window across system, history, retrieval, and output roles
• Give each RAG source a weighted share with guaranteed floors
• Reserve output tokens before splitting the remainder

Avoid for

• Measuring how many tokens text actually needs
• Non-integer or probabilistic allocation

Related skills

Composable with

Safety & privacy

• Runs without calling any external service.
• Your payloads are never stored permanently.
• Telemetry captures metadata only — never your payload contents.
• We never train models on your payloads.
• No human reviews your payloads.

Tags