Routis

Adaptive execution routing for AI coding workflows.

Why Routis | Before / After | Run Routis | How it works | Decision Trace | Policy

Routis is a local routing and audit layer for AI coding workflows.

It adds a deliberate routing step before execution. Given a task, Routis classifies the request, reads repository signals, applies local policy, selects an execution profile, and explains why that route was chosen.

The point is simple: AI-assisted development should not use the same execution path for every task. Small edits should stay lightweight. Risky changes should receive stronger handling. Model choice, reasoning depth, and command shape should be explicit instead of hidden in repeated shell habits.

Why Routis

AI coding tasks do not all deserve the same execution depth. A typo fix, a focused implementation task, and a risky debugging session should not be routed through identical model and reasoning settings.

Routis makes that decision explicit:

• Classify the task into a clear execution profile.
• Read repository context such as branch, changed files, manifests, tests, workflows, and sensitive areas.
• Apply local policy so model and reasoning choices stay reviewable.
• Explain the route with a compact Decision Trace.
• Keep local state privacy-aware by avoiding raw task text in traces and sessions by default.

Before / After

Without Routis, developers often choose model, reasoning depth, and command shape manually for every AI coding task. Small and risky changes can end up using the same execution path because the routing decision lives in memory, habit, or ad-hoc shell commands.

With Routis, the routing decision becomes a local, inspectable step before the AI agent runs.

Example route:

routis route --explain "debug auth flow"

selected: deep / gpt-5.5 / high
intent: debug
area: auth
risk: high

Routis Decision Trace
├─ Input Analysis
│  ├─ Intent: debug
│  ├─ Area: auth
│  └─ Risk: high
└─ Route Decision
   ├─ Selected profile: deep
   ├─ Model: gpt-5.5
   └─ Reasoning: high

Instead of relying on shell habits, Routis gives the route, the reason, and the selected execution profile before work begins.

Run Routis

Install from crates.io:

cargo install routis

Install from source:

git clone https://github.com/alenisaw/routis.git
cd routis
cargo install --path .

Open the interactive terminal UI:

routis

Preview a route from the CLI:

routis route "debug auth flow"

Print the route explanation tree:

routis route --explain "debug auth flow"

Inspect recent local traces:

routis traces
routis traces --latest

Cargo installs the binary into Cargo's bin directory, usually ~/.cargo/bin on Linux/macOS and %USERPROFILE%\.cargo\bin on Windows. Make sure that directory is on PATH.

How it works

Routis combines task classification, repository context, local policy, and execution profiles.

Profiles are intentionally simple:

Decision Trace

Routis can store routing decisions as local JSONL traces under:

~/.routis/traces/<session_id>.jsonl

Decision Trace is designed for local auditability: it records the selected route, matched signals, repository facts, policy source, selected model, and reasoning level.

By default, Routis avoids raw task text in trace and session records. CLI traces store an HMAC-SHA256 task hash using a per-install local secret at:

~/.routis/secret

Runtime files live under ~/.routis by default. Set ROUTIS_HOME to override this location.

Policy

Routis behavior is controlled through local policy.

Default policy path:

~/.routis/policies/default.yaml

Policy rules keep routing behavior local, reviewable, and project-specific. The default policy can be replaced with a custom YAML file when a repository needs different routing rules.

License

Routis is released under the Apache-2.0 License. See LICENSE.