Platform
How WaveAssist works.
Deterministic code, wrapped around AI intelligence. Your coding agent writes the pipeline, WaveAssist runs it the same way every time.
Lifecycle
From plain English to a scheduled agent.
Five stages, one MCP connection. You describe it, your agent builds it, WaveAssist proves it before it goes live.
connectyour MCP host
describein plain english
runon the cloud
verifyit works
deployon your schedule
The wrapped call
Deterministic output from a model.
call_llm takes a response_model. The model fills the content, the schema locks the shape.
review = waveassist.call_llm(
prompt=review_prompt(pr),
response_model=Review, # JSON schema
)prompt + input
model
JSON schema
same shape, every run
State across runs
Each run resumes from the last.
fetch_data reads what the last run stored, store_data writes what the next needs.
/agent · clickup-monday-summary
07:00run #4193dur 1.9scost $0.004passed
mon 07:00run #4181dur 2.1scost $0.004passed
nextscheduled mon 07:00repeats weeklyscheduled
The verify gate
It arms only after a passing run.
Every deploy does a real run on live infra. Pass, the schedule arms. Fail, it holds.
dry run
on live infra
pass → arms the schedule
fail → holds, stays off
Determinism vs re-run
The same answer, not a new one.
Re-running the prompt
· 4 items, upbeat tone
· 3 items, terse
· 5 items, formal
Full inference every tick. The shape drifts.
WaveAssist
Review{ summary, risk, comments }
Review{ summary, risk, comments }
Review{ summary, risk, comments }
Schema-locked. The shape holds.
Architecture
One connection, three layers.
Your editorClaude Code, Cursor, any MCP host
WaveAssist runtimecodified pipeline · wrapped model calls · scheduler · state
Your toolsGitHub, ClickUp, email, any API