One-line proxy from MyDream Labs helps cut back AI coding prices, wait time, and generated code quantity whereas protecting present supplier credentials.
AI coding brokers and agentic workflows are transferring extra developer work onto usage-based mannequin APIs, making output-token waste a direct price challenge for people and groups. MyDream Labs right this moment launched Distill, a drop-in optimization proxy for Claude Code and OpenAI Codex. In a 16-task inner benchmark, Distill measured reductions of as much as 50.6% in output-token use, as much as 55.3% in wait time, and as much as 30.7% in generated code quantity. Common reductions and full methodology are revealed on the Distill benchmark web page.
Waste from vibe coding compounds quick — each pointless token is cash now, each pointless line is upkeep price later.”
— Viacheslav Bogdanov, Founder, MyDream Labs
As extra AI coding instruments transfer towards usage-based billing, unbiased builders and small groups have much less room for inefficient agent output. For builders working brokers on private initiatives or shopper work, the associated fee lands immediately on a private or small-business card.
The economics of LLM APIs compound the problem: output tokens price builders 5x extra per token than enter. The fashions behind right this moment’s coding brokers have been initially constructed for conversational AI — the place thorough, multi-angle solutions are precisely what customers need. That coaching is the fitting habits for a chat interface and the mistaken habits for a coding agent, which wants one efficient answer executed cleanly, not a call tree. Distill optimizes round each: the verbosity that comes from chat-model coaching and the price of output tokens.
Additionally Learn: AiThority Interview with Matej Bukovinski, Chief Know-how Officer at Nutrient
Distill sits between the coding agent and the mannequin supplier, supporting each the Anthropic and OpenAI API codecs. Not like model-switching instruments or agent replacements, Distill works transparently inside the prevailing Claude Code or Codex workflow with no adjustments to how the agent or mannequin is used. Builders join utilizing their present Anthropic or OpenAI account — requests are billed precisely as if the agent known as the supplier immediately, with out request logging or vendor lock-in.
The impact compounds throughout periods: every activity that generates much less code leaves the following session with a leaner codebase, a quicker agent, and a smaller invoice.
Public beta highlights:
– Drop-in setup: one configuration line for Claude Code or Codex
– Works by way of your present Anthropic or OpenAI account — no separate credentials to handle
– No request logging — code and prompts aren’t retained by Distill
– $4/month for people, $40/month for groups
– 3-day f*********
“Waste from vibe coding compounds quick — each pointless token is cash now, each pointless line is upkeep price later. Distill was how I mounted my very own AI coding spend with out altering how I work,” stated Viacheslav Bogdanov, founding father of MyDream Labs.
“Earlier than Distill, I used to be continually asking the agent to simplify — lower this operate, don’t over-engineer, keep targeted. With Distill the code comes out clear. I barely should ask anymore,” stated one early beta tester.
Full inner outcomes and benchmark methodology are revealed on the Distill benchmark web page. Till a reproducible public benchmark is obtainable, probably the most correct check is the developer’s personal codebase and duties. Outcomes range by activity complexity; longer and extra advanced duties have a tendency to point out bigger features.
The general public beta is obtainable right this moment at https://distill.codes, with a 3-day trial constructed for side-by-side comparability on the developer’s personal work.
Additionally Learn: AI programs – Interoperable AI programs: Connecting fashions throughout platforms
[To share your insights with us, please write to psen@itechseries.com ]
