Context

Context controls what data the AI can see when processing a request. ThinkLang provides with context to include data and without context to exclude it.

with context: variable

Pass a single variable as context:

let email = "Congratulations! You've won a FREE iPhone!"

let result = think<Classification>("Classify this email")
  with context: email

The variable name and value are both sent to the AI, so the AI knows it is looking at an email.

with context: block

Pass multiple variables using a block:

let sourceText = "The quick brown fox jumps over the lazy dog"
let targetLang = "French"

let result = think<Translation>("Translate the source text to the target language")
  with context: {
    sourceText,
    targetLang,
  }

Each variable name becomes a labeled key in the context, helping the AI understand the role of each piece of data.

Chaining Context

You can pass AI results as context to subsequent calls:

let entities = think<Entities>("Extract technologies and challenges")
  with context: report

let analysis = think<Analysis>("Analyze the landscape based on these entities")
  with context: {
    entities,
    report,
  }

This enables multi-step workflows where each AI call builds on the previous one.

without context

Use without context to exclude sensitive data from being sent to the AI. This is critical for privacy -- you can include a broad context while stripping out specific fields:

let profile = "John Doe, age 30, interests: hiking, photography"
let sensitiveData = "SSN: 123-45-6789, Credit Card: 4111-1111-1111-1111"

let recommendation = think<Recommendation>("Suggest products based on user interests")
  with context: {
    profile,
    sensitiveData,
  }
  without context: sensitiveData

In this example, sensitiveData is excluded before the context is sent to the AI. The AI only sees profile.

Context Truncation

ThinkLang automatically truncates context that exceeds the model's token limit. Large string values are trimmed with a [truncated] marker, and a warning is emitted:

[ThinkLang] Context exceeds token limit (~120000 tokens). Truncating...

Context entries are sorted by size, so smaller entries are preserved in full before larger ones are truncated.

Best Practices

• Name variables descriptively. The variable name is sent to the AI as a label: sourceText is more informative than s.
• Include only relevant data. Smaller, focused context produces better results and costs less.
• Use without context for sensitive fields. Never send PII, credentials, or secrets to the AI.
• Chain context across calls. Build up understanding incrementally rather than dumping everything into one call.

Using from JS/TS

In the library API, context is passed via the context option and exclusions via withoutKeys. See the Library Core Functions guide for details.

import { think } from "thinklang";

// Pass context
const result = await think<{ category: string }>({
  prompt: "Classify this email",
  jsonSchema: { type: "object", properties: { category: { type: "string" } }, required: ["category"] },
  context: { email: "Congratulations! You've won!", subject: "Prize notification" },
});

// Exclude keys from context
const safe = await think<{ recommendation: string }>({
  prompt: "Suggest products based on interests",
  jsonSchema: { type: "object", properties: { recommendation: { type: "string" } }, required: ["recommendation"] },
  context: { profile, sensitiveData },
  withoutKeys: ["sensitiveData"],
});