ai-context-window-fit Edge Function — Ai

ai-context-window-fit — Trim conversation history to fit a model's context window

Removes messages from a conversation array to fit within a token budget, using configurable strategies to decide which messages to drop — no LLM API calls required.

---

API

POST /api/ai-context-window-fit

Request body

Field	Type	Required	Default	Description
messages	{role: string, content: string}[]	✅	—	Conversation messages
maxTokens	number	✅	—	Maximum token budget
model	"gpt-4" | "gpt-3.5-turbo" | "claude-3" | "claude-2" | "gemini-pro"	❌	"gpt-4"	Model for window ceiling
strategy	"trim-oldest" | "trim-middle" | "summarize-hint"	❌	"trim-oldest"	Trimming strategy
reserveTokens	number	❌	0	Tokens reserved for completion

Success response (200)

{
  "success": true,
  "data": {
    "messages": [{ "role": "system", "content": "..." }, { "role": "user", "content": "..." }],
    "estimatedTokens": 1240,
    "removedCount": 5,
    "fitsInWindow": true
  }
}

Error responses

Code	HTTP	When
INVALID_INPUT	400	Missing required field or wrong type
INTERNAL_ERROR	500	reserveTokens leaves no room

---

Usage

cURL

curl -X POST "$FUNCTION_URL" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Tell me about AI."},
      {"role": "assistant", "content": "AI is..."}
    ],
    "maxTokens": 4000,
    "strategy": "trim-oldest",
    "reserveTokens": 500
  }'

TypeScript / JavaScript (HTTP)

const response = await fetch(FUNCTION_URL, {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ messages, maxTokens: 8000, strategy: 'trim-middle' }),
});
const { data } = await response.json();
const safeMessages = data.messages;

Direct import (Node / Bun / Deno)

import { aiContextWindowFit } from '@aerostack/functions/ai-context-window-fit';

const { messages, estimatedTokens } = aiContextWindowFit({
  messages: conversationHistory,
  maxTokens: 16000,
  reserveTokens: 2000,
});

---

Use Cases

• Trimming a growing chat history before each LLM API call to avoid context_length_exceeded errors.
• Building multi-turn agents that need to maintain recent context while staying within token limits.
• Estimating token usage of a conversation array without making any API calls.

---

Notes

• Token estimation uses ~4 chars/token + 4 tokens per message for role/formatting overhead.
• trim-oldest never removes the first system message.
• summarize-hint inserts a system note informing the model that earlier context was trimmed.
• maxTokens is capped at the model's absolute context window even if you pass a higher value.