Run deterministic AI agents
in the cloud
Describe a job in plain English. Your coding agent builds it, WaveAssist runs it on your schedule.
- Deterministic by contract
- Built by your coding agent
- Any MCP host
Build your first agent from your editor.
Three steps, no docs required. Your first agent in a few minutes.
Get your MCP token
Sign up at app.waveassist.io and copy your MCP token. $2 of runtime credit is included.
Connect your editor
One command for Claude Code, one JSON block for Cursor, one URL for any other MCP host.
Ask for the job
Describe a job in plain English. Your agent designs it, tests it live, and schedules it once it passes.
claude mcp add --transport http waveassist \ https://mcp.waveassist.ai/mcp \ --header "Authorization: Bearer YOUR_TOKEN"
"Using WaveAssist, build an agent that reads my ClickUp and emails me a Monday morning summary."
Deterministic by construction.
A codified pipeline
Fetch, transform, post, and persist run as deterministic Python. Same flow every run.
One wrapped call
call_llm fills content into a fixed schema. Structured JSON out, no drift.
State that persists
fetch_data and store_data carry state, so each run resumes from the last.
from pydantic import BaseModel
import waveassist
class Review(BaseModel):
summary: str
risk: str
comments: list[str]
def run():
pr = waveassist.fetch_data("pull_request")
# model fills content, shape is locked
review = waveassist.call_llm(
prompt=review_prompt(pr),
response_model=Review,
)
post_comment(pr, review)The model fills the content. response_model locks the shape.
Millions of runs, deterministic every time.
Deterministic code, wrapped around AI intelligence.
Deterministic by contract
Every model call is locked to a JSON schema, so every run returns the same structure.
Set up from your editor, hosted by us
Describe the job in plain English and your coding agent ships it over one MCP connection. We host and run it.
A fraction of the cost
Codified code runs for a fraction of a cent per run. Full-inference schedulers pay every tick, a 10x to 100x gap.
Verified before it arms
Every agent proves itself on a live run before its schedule arms.
Against the two tools you would otherwise reach for.
Editor schedulers re-run the whole prompt every tick. Durable infra makes you babysit the harness. WaveAssist is both at once.
The public repos at github.com/WaveAssist stay up as examples of what the runtime runs.
Point your coding agent
at the runtime.
Connect the MCP and let it build, test, and schedule your first agent. $2 of runtime credit is on us.