brevitas-systems 0.1.0

Token-efficient multi-agent pipeline optimization

brevitas-systems

Token-efficient multi-agent pipeline optimization.

pip install brevitas-systems

from brevitas import optimize
from my_pipeline import architect, builder, reviewer

pipeline = optimize([architect, builder, reviewer])
result = pipeline.run(task)
# ↳ 59% fewer tokens. 47% lower cost. 99% quality parity.

Setup

Generate an API key at brevitas.systems/dashboard, then either:

export BREVITAS_API_KEY=bvt_your_key_here

or pass it directly:

from brevitas import optimize, configure

configure(api_key="bvt_your_key_here")
pipeline = optimize([architect, builder, reviewer])

How it works

Brevitas wraps your existing agent pipeline and applies five complementary token-reduction tactics on every turn:

Tactic	What it does
Task-aware routing	Sends each task to the smallest model that can handle it
Smart context pruning	Drops low-relevance context before it hits the prompt
Adaptive semantic sampling	Scores context chunks by relevance, recency, and novelty
Communication compression	Strips redundancy from inter-agent messages
Shared memory + delta payloads	Sends only what changed since the last turn

API reference

optimize(agents, *, api_key=None, **kwargs) → OptimizedPipeline

Wrap a list of agents in the Brevitas optimization layer.

Param	Type	Default	Description
agents	list	required	Agent callables. Each receives the previous agent's output.
api_key	str	env var	Your bvt_ prefixed API key.
quality_floor	float	0.98	Min quality proxy before the pipeline stops compressing.
savings_target	float	59.0	Token savings % to target per turn.
compression_level	int	2	Message compression aggressiveness (1–3).
prune_budget	int	5	Max context items retained per turn.

OptimizedPipeline.run(task, **kwargs) → PipelineResult

Run a task through the pipeline. Returns a PipelineResult with:

• savings_pct — tokens saved vs. unoptimized baseline
• model_response — concatenated agent outputs
• baseline_tokens / optimized_tokens — raw counts
• quality_proxy — estimated quality retention (0–1)
• debug — internal metrics (compression, sampling, pruning)