inferroute 0.3.1__tar.gz

inferroute-0.3.1/.github/workflows/release.yml

@@ -0,0 +1,56 @@
+name: Publish to PyPI
+
+# Publishes `inferroute` to PyPI via Trusted Publishing (OIDC) — no API token
+# is stored in the repo. Fires when a GitHub Release is published (the normal
+# path), and also exposes a manual button for re-runs.
+#
+# One-time setup on PyPI (Account → Publishing → Add a pending publisher):
+#   PyPI project name : inferroute
+#   Owner             : InferRoute
+#   Repository name   : inferroute
+#   Workflow filename : release.yml
+#   Environment name  : pypi
+# The environment name MUST match the `environment:` block below.
+
+on:
+  release:
+    types: [published]
+  workflow_dispatch:
+
+jobs:
+  build:
+    name: Build distributions
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - name: Build sdist and wheel
+        run: |
+          python -m pip install --upgrade build
+          python -m build
+      - name: Check metadata
+        run: |
+          python -m pip install --upgrade twine
+          python -m twine check dist/*
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+  publish:
+    name: Publish to PyPI
+    needs: build
+    runs-on: ubuntu-latest
+    environment:
+      name: pypi
+      url: https://pypi.org/p/inferroute
+    permissions:
+      id-token: write  # required for Trusted Publishing (OIDC)
+    steps:
+      - uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+      - uses: pypa/gh-action-pypi-publish@release/v1

inferroute-0.3.1/.gitignore

@@ -0,0 +1,11 @@
+__pycache__/
+*.py[cod]
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+*.egg-info/
+build/
+dist/
+.venv/
+.env
+.envrc

inferroute-0.3.1/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Henry Declety / inferroute
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

inferroute-0.3.1/PKG-INFO

@@ -0,0 +1,137 @@
+Metadata-Version: 2.4
+Name: inferroute
+Version: 0.3.1
+Summary: Launch Claude Code through inferroute — typed less, routed smarter
+Project-URL: Homepage, https://inferroute.ai
+Project-URL: Source, https://github.com/InferRoute/inferroute
+Project-URL: Issues, https://github.com/InferRoute/inferroute/issues
+Author-email: Henry Declety <henry@inferroute.ai>
+License-Expression: MIT
+License-File: LICENSE
+Keywords: ai,claude,claude-code,gateway,llm,router
+Classifier: Development Status :: 4 - Beta
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Topic :: Software Development
+Requires-Python: >=3.10
+Requires-Dist: httpx<1.0,>=0.27
+Requires-Dist: textual<2.0,>=0.78
+Provides-Extra: local
+Requires-Dist: click>=8.1; extra == 'local'
+Requires-Dist: fastapi>=0.111; extra == 'local'
+Requires-Dist: httpx>=0.27; extra == 'local'
+Requires-Dist: uvicorn[standard]>=0.30; extra == 'local'
+Description-Content-Type: text/markdown
+
+# ir — Claude Code through inferroute
+
+Tiny launcher: one command (`ir`) spawns `claude` with the model you pick
+and your inferroute key, so you don't have to set env vars every time.
+**Your normal `claude` is never touched** — `ir` only configures the
+subprocess it spawns.
+
+You choose the model per session — there's no auto-routing. Optionally, an
+on-device recorder logs your choices privately (see [Local recording](#local-recording-optional)).
+
+## Install
+
+```bash
+pipx install inferroute
+ir login
+```
+
+That's it. `ir login` saves your API key to `~/.config/inferroute/credentials`
+(mode 600) and verifies it against the inferroute API. Update any time with
+`pipx upgrade inferroute`.
+
+Sign up for a key at https://inferroute.ai if you don't have one.
+
+## Commands
+
+```
+ir                  open the model picker, then launch (same as `ir choose`)
+ir --model NAME     pin a model — short alias or canonical id; any claude flag passes through
+ir choose           interactive picker — pick a model, then launch
+ir anthropic        escape hatch — plain Claude, your own setup
+ir status           personal usage view (TUI)
+ir login            save / refresh your inferroute API key
+ir help             one-screen explainer
+
+ir add recording    turn on private on-device recording (see below)
+ir data show        what's been recorded (counts, models, size)
+ir data export DIR  copy the metadata layer  ·  ir data wipe  delete it all
+ir remove recording stop recording
+```
+
+`NAME` is a short alias (`minimax`, `minimax-m3`, `kimi`, `glm`) or any
+canonical id (e.g. `claude-opus-4-8`), which passes through verbatim.
+
+Every command bakes in `--dangerously-skip-permissions`, which is the
+right default for an agentic workflow. If you want the prompts back,
+just use plain `claude`.
+
+## What does each command do under the hood?
+
+```
+ir --model NAME      →  ANTHROPIC_BASE_URL=<recorder daemon if running,         \\
+                                            else https://api.inferroute.ai>      \\
+                        ANTHROPIC_AUTH_TOKEN=$saved_key                          \\
+                        exec claude --dangerously-skip-permissions              \\
+                                    --model <resolved-id>
+ir anthropic         →  exec claude --dangerously-skip-permissions
+                        (no env mutation — pure pass-through)
+```
+
+The env vars are scoped to the spawned `claude` process. Your shell
+doesn't see them. When the local recorder daemon is running it sits in front
+of `api.inferroute.ai` — it records locally, then forwards there (the cloud
+still does its own backend fallbacks).
+
+## Local recording (optional)
+
+`ir add recording` runs a small daemon on `localhost:5005` that records — **on
+this machine only** — which model you pick per task and how the turn went, so
+you (or a future personal router) can learn your preferences. It is **never
+uploaded; we never see it.**
+
+- Default is `full` — choices, outcomes, **and** prompt text. The prompt text is
+  what actually lets it learn your preferences, and it never leaves this machine.
+  Pick `minimal` at install for choices + outcomes only (no prompt text), which is
+  lighter but can't train a personal router.
+- Inspect it any time with `ir data show`, copy the metadata layer with
+  `ir data export DIR` (never raw prompt text), or delete everything with
+  `ir data wipe`.
+- It lives under `~/.inferroute`. Remove the daemon with `ir remove recording`
+  (`--purge` also deletes the recorded data).
+
+## Where your config lives
+
+| Path | Contents |
+|---|---|
+| `~/.config/inferroute/credentials` | Your inferroute API key + base URL (mode 600). Edit by hand or re-run `ir login`. |
+
+## Privacy
+
+The CLI talks to:
+- `https://api.inferroute.ai` for your traffic (everything except `ir anthropic`)
+- `https://inferroute.ai` only on `ir help` — to print the signup URL
+
+It does **not** phone home, no telemetry. If you enable `ir add recording`,
+that data is written **only** to `~/.inferroute` on your machine and is never
+uploaded — inspect or delete it any time with `ir data show` / `ir data wipe`.
+Whatever your `claude --model …` sends to whatever upstream is on you.
+
+`ir anthropic` is literally `exec claude` plus one flag — the inferroute
+service sees nothing.
+
+## Source
+
+https://github.com/InferRoute/inferroute
+
+Issues / feature requests / PRs welcome.
+
+## License
+
+MIT — see [LICENSE](LICENSE).

inferroute-0.3.1/README.md

@@ -0,0 +1,110 @@
+# ir — Claude Code through inferroute
+
+Tiny launcher: one command (`ir`) spawns `claude` with the model you pick
+and your inferroute key, so you don't have to set env vars every time.
+**Your normal `claude` is never touched** — `ir` only configures the
+subprocess it spawns.
+
+You choose the model per session — there's no auto-routing. Optionally, an
+on-device recorder logs your choices privately (see [Local recording](#local-recording-optional)).
+
+## Install
+
+```bash
+pipx install inferroute
+ir login
+```
+
+That's it. `ir login` saves your API key to `~/.config/inferroute/credentials`
+(mode 600) and verifies it against the inferroute API. Update any time with
+`pipx upgrade inferroute`.
+
+Sign up for a key at https://inferroute.ai if you don't have one.
+
+## Commands
+
+```
+ir                  open the model picker, then launch (same as `ir choose`)
+ir --model NAME     pin a model — short alias or canonical id; any claude flag passes through
+ir choose           interactive picker — pick a model, then launch
+ir anthropic        escape hatch — plain Claude, your own setup
+ir status           personal usage view (TUI)
+ir login            save / refresh your inferroute API key
+ir help             one-screen explainer
+
+ir add recording    turn on private on-device recording (see below)
+ir data show        what's been recorded (counts, models, size)
+ir data export DIR  copy the metadata layer  ·  ir data wipe  delete it all
+ir remove recording stop recording
+```
+
+`NAME` is a short alias (`minimax`, `minimax-m3`, `kimi`, `glm`) or any
+canonical id (e.g. `claude-opus-4-8`), which passes through verbatim.
+
+Every command bakes in `--dangerously-skip-permissions`, which is the
+right default for an agentic workflow. If you want the prompts back,
+just use plain `claude`.
+
+## What does each command do under the hood?
+
+```
+ir --model NAME      →  ANTHROPIC_BASE_URL=<recorder daemon if running,         \\
+                                            else https://api.inferroute.ai>      \\
+                        ANTHROPIC_AUTH_TOKEN=$saved_key                          \\
+                        exec claude --dangerously-skip-permissions              \\
+                                    --model <resolved-id>
+ir anthropic         →  exec claude --dangerously-skip-permissions
+                        (no env mutation — pure pass-through)
+```
+
+The env vars are scoped to the spawned `claude` process. Your shell
+doesn't see them. When the local recorder daemon is running it sits in front
+of `api.inferroute.ai` — it records locally, then forwards there (the cloud
+still does its own backend fallbacks).
+
+## Local recording (optional)
+
+`ir add recording` runs a small daemon on `localhost:5005` that records — **on
+this machine only** — which model you pick per task and how the turn went, so
+you (or a future personal router) can learn your preferences. It is **never
+uploaded; we never see it.**
+
+- Default is `full` — choices, outcomes, **and** prompt text. The prompt text is
+  what actually lets it learn your preferences, and it never leaves this machine.
+  Pick `minimal` at install for choices + outcomes only (no prompt text), which is
+  lighter but can't train a personal router.
+- Inspect it any time with `ir data show`, copy the metadata layer with
+  `ir data export DIR` (never raw prompt text), or delete everything with
+  `ir data wipe`.
+- It lives under `~/.inferroute`. Remove the daemon with `ir remove recording`
+  (`--purge` also deletes the recorded data).
+
+## Where your config lives
+
+| Path | Contents |
+|---|---|
+| `~/.config/inferroute/credentials` | Your inferroute API key + base URL (mode 600). Edit by hand or re-run `ir login`. |
+
+## Privacy
+
+The CLI talks to:
+- `https://api.inferroute.ai` for your traffic (everything except `ir anthropic`)
+- `https://inferroute.ai` only on `ir help` — to print the signup URL
+
+It does **not** phone home, no telemetry. If you enable `ir add recording`,
+that data is written **only** to `~/.inferroute` on your machine and is never
+uploaded — inspect or delete it any time with `ir data show` / `ir data wipe`.
+Whatever your `claude --model …` sends to whatever upstream is on you.
+
+`ir anthropic` is literally `exec claude` plus one flag — the inferroute
+service sees nothing.
+
+## Source
+
+https://github.com/InferRoute/inferroute
+
+Issues / feature requests / PRs welcome.
+
+## License
+
+MIT — see [LICENSE](LICENSE).

inferroute-0.3.1/docs/RELEASING.md

@@ -0,0 +1,120 @@
+# Releasing inferroute
+
+Two artifact families need to ship together for a clean release: the Python
+package (PyPI) and the classifier model bundle (GitHub Releases). The CLI's
+default `classifier_bootstrap_url` points at the GitHub release's "latest"
+download path, so users running `ir add local-routing` get whichever model
+was attached to the most recent tag.
+
+## 1. Cut the Python release
+
+```bash
+# Bump the version in pyproject.toml, commit, tag
+vim pyproject.toml                            # e.g. 0.1.0 → 0.2.0
+git commit -am "release v0.2.0"
+git tag v0.2.0
+git push origin main --tags
+
+# Build wheels
+rm -rf dist/
+python -m build
+
+# Upload to PyPI (uses your ~/.pypirc or twine env vars)
+python -m twine upload dist/*
+```
+
+## 2. Build the classifier bundle
+
+The bundle is the contents of `~/.inferroute/models/classifier-v0/` packed
+as individual files + a `classifier-v0-manifest.json` describing them.
+
+```bash
+# From inferroute-local-experiments — the canonical source of the trained model
+cd ~/workspaces/inferroute/inferroute-local-experiments
+MODEL_DIR=out/classifier-v0-longer
+BUNDLE_DIR=/tmp/classifier-v0-bundle-$(date +%Y%m%d)
+mkdir -p $BUNDLE_DIR/onnx
+
+# Copy the four required files (and their data sidecar if present)
+cp $MODEL_DIR/onnx/model.onnx $BUNDLE_DIR/onnx/
+[ -f $MODEL_DIR/onnx/model.onnx.data ] && cp $MODEL_DIR/onnx/model.onnx.data $BUNDLE_DIR/onnx/
+cp $MODEL_DIR/tokenizer.json $BUNDLE_DIR/
+cp $MODEL_DIR/calibration.json $BUNDLE_DIR/
+cp $MODEL_DIR/label_to_int.json $BUNDLE_DIR/
+
+# Generate the manifest (sha256 each file, point url at the release tag)
+TAG=v0.2.0   # match the Python release tag
+.venv/bin/python - <<EOF
+import hashlib, json
+from pathlib import Path
+base = Path("$BUNDLE_DIR")
+release_url = "https://github.com/InferRoute/inferroute/releases/download/$TAG"
+files = []
+for rel in ["onnx/model.onnx", "tokenizer.json", "calibration.json", "label_to_int.json"]:
+    p = base / rel
+    if not p.exists():
+        continue
+    sha = hashlib.sha256(p.read_bytes()).hexdigest()
+    files.append({"path": rel, "url": f"{release_url}/{p.name}", "sha256": sha})
+# Optional sidecar (large model weights)
+for sidecar in base.glob("onnx/model.onnx.data"):
+    sha = hashlib.sha256(sidecar.read_bytes()).hexdigest()
+    files.append({"path": f"onnx/{sidecar.name}", "url": f"{release_url}/{sidecar.name}", "sha256": sha})
+manifest = {"version": "$TAG", "files": files}
+(base / "classifier-v0-manifest.json").write_text(json.dumps(manifest, indent=2))
+print(json.dumps(manifest, indent=2))
+EOF
+```
+
+## 3. Attach to the GitHub Release
+
+Upload these as release assets to the matching tag (`v0.2.0`):
+
+- `classifier-v0-manifest.json`   ← the daemon's entry point; must be named exactly this
+- `model.onnx`                    ← rename so the URLs in the manifest match
+- `model.onnx.data`               ← if your export uses external weights
+- `tokenizer.json`
+- `calibration.json`
+- `label_to_int.json`
+
+Easiest via `gh`:
+
+```bash
+gh release create $TAG \
+  --title "$TAG" \
+  --notes "$(git log --oneline $(git describe --tags --abbrev=0 HEAD^)..HEAD)" \
+  $BUNDLE_DIR/classifier-v0-manifest.json \
+  $BUNDLE_DIR/onnx/model.onnx \
+  $BUNDLE_DIR/onnx/model.onnx.data \
+  $BUNDLE_DIR/tokenizer.json \
+  $BUNDLE_DIR/calibration.json \
+  $BUNDLE_DIR/label_to_int.json
+```
+
+The daemon's default `classifier_bootstrap_url` uses `/releases/latest/download/`,
+so once this release is published, every user running `ir add local-routing`
+gets these files automatically.
+
+## 4. Smoke-test the published release
+
+```bash
+# In a clean env that doesn't have the model on disk
+pip install --upgrade inferroute
+rm -rf ~/.inferroute/models/classifier-v0
+ir add local-routing --no-service --no-shell-edit --yes
+# Should print: ✓ Installed model version v0.2.0 at ~/.inferroute/models/classifier-v0
+ls ~/.inferroute/models/classifier-v0/
+```
+
+## Common slip-ups
+
+- **Filename mismatch**: the manifest's `url` field must match the actual
+  filename you uploaded. `model.onnx.data` in the manifest doesn't help if
+  you uploaded it as `weights.bin`.
+- **Tag drift**: if you bump the Python version but forget to tag, the next
+  `ir add local-routing` still pulls the OLD model from `/releases/latest/`.
+  Always tag both halves at the same version.
+- **Pre-release tag**: GitHub's `/releases/latest/` ignores pre-releases. If
+  you ship a `v0.3.0-rc1` and don't mark it as a full release, users still
+  see `v0.2.0`. Use this on purpose during testing, but remember to flip
+  the "Set as the latest release" toggle for the real ship.

inferroute-0.3.1/inferroute_cli/__init__.py

@@ -0,0 +1,8 @@
+"""inferroute CLI — launch Claude Code through inferroute with one command."""
+from importlib.metadata import PackageNotFoundError, version
+
+try:
+    # Single source of truth: the installed package version from pyproject.
+    __version__ = version("inferroute")
+except PackageNotFoundError:  # running from a source tree that isn't installed
+    __version__ = "0.0.0+dev"

inferroute-0.3.1/inferroute_cli/__main__.py

@@ -0,0 +1,7 @@
+"""Allow `python -m inferroute_cli`."""
+import sys
+
+from .main import main
+
+if __name__ == "__main__":
+    sys.exit(main())