Write the skill
Use natural language, references, and tools. Let technical and non-technical teammates shape what good looks like.
SKILL.md + references
An agent-skill compiler
Discover with skills.
Ship as code.
Skills let anyone on the team define a working automation in prose. Rote turns the parts that should repeat into typed, reviewable code that one engineer can maintain, without losing the judgment calls that still need AI.
compilation / ticket-triage● complete
SKILL.md
Read the ticket, verify the account,
route by severity, draft a response…
Read the ticket, verify the account,
route by severity, draft a response…
compiled into
verify_accountpure_function
route_severitypure_function
judge_ambiguityllm_judge
create_ticketexternal_call
run 01 ✓
run 02 ✓
run 03 ✓
Apache 2.0open source CLI
6 runtimesfrom Python to Cloudflare
5 node kindsone portable typed DAG
pre-1.0built in the open
From exploration to operation
Keep the magic.
Remove the guesswork.
Agents are excellent at discovery. Repeating a discovered process with another open-ended agent loop is expensive, slow, and hard to test. Rote finds the boundary.
Extract the routine
Rote classifies each step, lifts deterministic logic into modules, and preserves bounded AI and human decisions.
pipeline.yaml + typed modules
Run it as code
One engineer reviews, tests, versions, and deploys what the whole team authors. The skill stays the editable source of truth.
DBOS · Temporal · Cloudflare · more
A real command, not a new platform
Skill in.
Workflow out.
Point Rote at an Anthropic-style skill. The output is runtime-agnostic IR, extracted modules, typed judge signatures, and deployable runtime code.
# run without installing
$ uvx --from rote-cli rote compile ./my-skill \
--runtime cloudflare --out ./compiled
✓ analyzed 19 steps
✓ extracted 15 deterministic nodes
✓ preserved 4 bounded decisions
✓ emitted Cloudflare Workflow
$ cd compiled && wrangler deploy
$ uvx --from rote-cli rote compile ./my-skill \
--runtime cloudflare --out ./compiled
✓ analyzed 19 steps
✓ extracted 15 deterministic nodes
✓ preserved 4 bounded decisions
✓ emitted Cloudflare Workflow
$ cd compiled && wrangler deploy
One workflow model. Two ways to run it.
Own the stack.
Or skip the ops.
The open-source CLI defines the portable workflow. Cloud is the managed execution path, not a different, locked-down product.
Open source
Rote CLI
Compile locally, inspect every artifact, and deploy to infrastructure you control.
- Apache 2.0 licensed
- Portable typed DAG
- DBOS, Temporal, Python, Cloudflare, DBOS TS, Inngest
- Your models, data, and runtime
Managed execution
Rote Cloud
Deploy compiled pipelines to isolated, durable Cloudflare Workflows and measure what matters across real runs.
- Managed durable execution
- Measured speed and cost per run
- Repeated-run determinism reports
- Tenant-isolated pipeline artifacts
Why compile?
Repetition changes the economics.
Exploratory work should stay exploratory. When a skill has run twenty times and needs to run a thousand more unattended, code earns its keep.
Rote's own results depend on the workflow. These figures are from independent research comparing agentic and deterministic implementations at scale.
Read the research ↗57×fewer tokens at 1,000 transactions
450×lower median latency
100%reproducibility, vs 95% for direct inference at temperature 0
~40×lower TCO at 1M transactions / month
Direct answers
Common questions
Short answers to what teams ask before they compile their first skill.
What is Rote?
Rote is an open-source CLI, Apache-2.0 licensed and published as rote-cli on PyPI, that compiles a proven AI agent skill into a typed, deterministic workflow. It reads an Anthropic-style SKILL.md, classifies each step, moves fixed logic and tool orchestration into reviewable code, and calls a model only for the steps that genuinely require judgment.
Who is Rote for?
Teams where the people who know the procedure are not the people who write code. A non-technical teammate authors and edits the skill as prose, the compiler turns it into a typed workflow, and an engineer reviews the emitted diff instead of hand-translating another team's requirements; rote compile --update picks up later prose edits without starting over. Engineers automating their own work save the extraction and test scaffolding too, but the compounding value is maintaining the skills you did not write.
How does Rote reduce AI agent token costs?
It removes model calls rather than making them cheaper. A repeating agent spends tokens re-reading instructions, tool schemas, and history to re-derive a procedure it already established. Rote compiles that procedure into code, so a repeated run pays only for the steps still classified as needing judgment.
When should I compile a skill instead of leaving it as an agent?
Keep one-off exploration in an agent, which is what agents are good at. Compile a skill once the procedure is proven, repeats often, and needs lower cost, faster execution, regression tests, explicit approvals, or reliable retries.
Does Rote replace my agent framework or MCP?
No. A compiled workflow can still call authenticated MCP servers and retain bounded agent loops. Rote decides which parts of a process should stop being inference; your runtime and your integrations stay where they are.
Where does the compiled workflow run?
Anywhere you already run durable work. Rote emits DBOS, Temporal, Cloudflare Workflows, plain Python, DBOS TypeScript, and Inngest. Rote Cloud is an optional managed path for teams that would rather not operate the runtime themselves.