History Processor

chimera.context.history provides composable processors for transforming conversation history before sending it to the LLM.

Key Classes

Class	Description
HistoryProcessor	Abstract base class — implement process(messages)
TruncateProcessor	Keep only the last N messages
PruneProcessor	Replace old tool result content with [pruned]
CompressProcessor	Compress old messages into a summary, keep recent ones intact
CompositeProcessor	Chain multiple processors in sequence

Quick Start

from chimera.context.history import TruncateProcessor, PruneProcessor, CompositeProcessor

# Keep last 20 messages, prune old tool results beyond the 3 most recent
processor = CompositeProcessor([
    PruneProcessor(keep_last_n_results=3),
    TruncateProcessor(max_messages=20),
])

cleaned = processor.process(messages)

Processors

TruncateProcessor

Keeps only the last N messages, discarding everything older.

proc = TruncateProcessor(max_messages=15)

PruneProcessor

Replaces old tool result content with [pruned], preserving structure.

proc = PruneProcessor(keep_last_n_results=5)

CompressProcessor

Compresses old messages into a summary, keeping recent ones intact. Uses simple concatenation and truncation (no LLM call).

proc = CompressProcessor(keep_recent=5, max_summary_tokens=500)

CompositeProcessor

Chains multiple processors — each processor’s output becomes the next input.

proc = CompositeProcessor([PruneProcessor(), TruncateProcessor()])

Custom Processors

Subclass HistoryProcessor and implement process():

from chimera.context.history import HistoryProcessor

class DropSystemProcessor(HistoryProcessor):
    def process(self, messages):
        return [m for m in messages if m.role != "system"]

Related

• Compaction — threshold-based context compaction
• Focus Chain — token-budget context selection