AI Tooling Built for Humans

Human intent is all that matters

CodeSpeak is a next-generation programming system powered by LLMs

Alpha Preview

Try Intent Recovery

Recover the requirements buried in your agent sessions.

Upload your sessions

npx @codespeak/helper

Already uploaded?

Open Intent Studio

Try the CodeSpeak CLI

to generate code from specs

NewBlog:Intent Recovery: The specs you meant to write

Production-grade Systems

Long-term projects,

not just prototypes

Engineers building complex software

Not vibe-coders

Teams of Humans

Not just solopreneurs

Communication matters

Maintain Specs, Not Code

See CodeSpeak in action

CodeSpeak works in mixed projects where some code is written manually and some is generated from specs. Here's an example from the MarkItDown repository (forked)

Check out the step-by-step guide on mixed projects

Turning Code into Specs Coming Soon

CodeSpeak can take over some of the existing code and replace it with specs 5-10x smaller. Maintaining specs is a lot easier for humans.

Real-World Case Studies

We took real code from open-source projects and generated specs from it. Here's how it panned out:

Case Study	Code LOC^[1]	Spec LOC^[1]	Shrink Factor	Tests Passed
WebVTT subtitles support for yt-dlp (video downloader) View spec & code	255	38	6.7x	before: 1241/1242 after: 1278/1279 (37 tests added)
Italian SSN generator for Faker (python library for generating mock data) View spec & code	165^[2]	21	7.9x	before: 2216 after: 2229 (13 tests added)
Encoding auto-detection and normalization for beautifulsoup4 (Python library for parsing HTML and XML) View spec & code	826	141	5.9x	before: 889 after: 914 (25 tests added)
EML to .md converter for markitdown (Python library for converting anything to markdown) View spec & code	139	14	9.9x	before: 165 after: 192 (27 tests added)

^[1] When computing LOC, we strip blank lines and break long lines into many
^[2] List of Italian municipalities codes (~8000 LOC) is excluded