Quickstart

Build a working context pipeline in 30 seconds. All you need is a system prompt and a memory manager.

Minimal pipeline

from anchor import ContextPipeline, QueryBundle, MemoryManager

# Create pipeline with memory
memory = MemoryManager(conversation_tokens=4096)
pipeline = (
    ContextPipeline(max_tokens=8192)
    .with_memory(memory)
    .add_system_prompt("You are a helpful coding assistant.")
)

# Add conversation history
memory.add_user_message("Help me write a Python function")
memory.add_assistant_message("Sure! What should the function do?")

# Build context for the next query
result = pipeline.build(QueryBundle(query_str="It should sort a list"))
print(result.formatted_output)

What just happened?

1. MemoryManager stores conversation turns in a sliding window capped at 4 096 tokens.
2. ContextPipeline assembles a context window of up to 8 192 tokens, placing the system prompt first (highest priority), then conversation memory, then any retrieval results.
3. build() packs everything that fits into a ContextResult -- ready to send to any LLM.

[!TIP] String shorthand build() also accepts a plain string. These two calls are equivalent:

result = pipeline.build("It should sort a list")
result = pipeline.build(QueryBundle(query_str="It should sort a list"))

What's next?

This pipeline has no retrieval -- it only uses memory and a system prompt. To add semantic search, hybrid retrieval, token budgets, formatters, and more, continue to Your First Pipeline.