release: PyPI distribution name → fastcrest-tether (bare 'tether' is blocked on PyPI)

.github/workflows/doctor-smoke.yml

@@ -83,7 +83,7 @@ jobs:
           # exercises the LD_LIBRARY_PATH patch path without pulling tensorrt
           # (which has no Mac wheel and is huge on Linux). Doctor should still
           # exit 0 with TRT libs ⚠ NOT installed warnings.
-          /tmp/tether-doctor-venv/bin/python -m pip install "$(ls dist/tether-*.whl)[onnx]"
+          /tmp/tether-doctor-venv/bin/python -m pip install "$(ls dist/fastcrest_tether-*.whl)[onnx]"
 
       - name: Verify tether --version matches pyproject.toml
         shell: bash

GOALS.yaml

@@ -213,7 +213,7 @@ goals:
 
   - id: fresh-install-verified
     status: done
-    description: "`pip install 'tether[serve,gpu] @ git+https://...'` succeeds on a fresh python:3.12-slim container and `tether --help` runs. Catches dep resolution breakage, broken editable installs, missing files in the wheel. Regression-check every release."
+    description: "`pip install 'fastcrest-tether[serve,gpu] @ git+https://...'` succeeds on a fresh python:3.12-slim container and `tether --help` runs. Catches dep resolution breakage, broken editable installs, missing files in the wheel. Regression-check every release."
     check: "test -f tests/test_fresh_install.py && .venv/bin/python -m pytest tests/test_fresh_install.py -q 2>/dev/null"
     weight: 8
 
@@ -416,8 +416,8 @@ goals:
     weight: 10
 
   - id: serve-one-line-install
-    description: "pip install tether && tether serve <hf-id> works on a Jetson with no Docker, no Triton YAML, no NVIDIA Container Toolkit. The DX win that beats Triton even though Triton wins on raw features. Most customers will tolerate a 1.5x latency hit for a 60s install vs a 4-hour Triton-on-Jetson setup."
-    check: "test -f docs/quickstart_jetson.md && grep -q 'pip install tether' README.md 2>/dev/null"
+    description: "pip install fastcrest-tether && tether serve <hf-id> works on a Jetson with no Docker, no Triton YAML, no NVIDIA Container Toolkit. The DX win that beats Triton even though Triton wins on raw features. Most customers will tolerate a 1.5x latency hit for a 60s install vs a 4-hour Triton-on-Jetson setup."
+    check: "test -f docs/quickstart_jetson.md && grep -q 'pip install fastcrest-tether' README.md 2>/dev/null"
     weight: 7
 
   # ── DEFERRED LONG-TERM SERVE (Phase 4-5; do NOT build now) ────────

README.md

@@ -2,10 +2,10 @@
 
 > by [FastCrest](https://fastcrest.com) — deployment infrastructure for vision-language-action models.
 
-[![PyPI](https://img.shields.io/pypi/v/tether.svg)](https://pypi.org/project/tether/)
-[![Python](https://img.shields.io/pypi/pyversions/tether.svg)](https://pypi.org/project/tether/)
-[![License](https://img.shields.io/pypi/l/tether.svg)](https://github.com/FastCrest/tether/blob/main/LICENSE)
-[![Downloads](https://img.shields.io/pypi/dm/tether.svg)](https://pypi.org/project/tether/)
+[![PyPI](https://img.shields.io/pypi/v/fastcrest-tether.svg)](https://pypi.org/project/fastcrest-tether/)
+[![Python](https://img.shields.io/pypi/pyversions/fastcrest-tether.svg)](https://pypi.org/project/fastcrest-tether/)
+[![License](https://img.shields.io/pypi/l/fastcrest-tether.svg)](https://github.com/FastCrest/tether/blob/main/LICENSE)
+[![Downloads](https://img.shields.io/pypi/dm/fastcrest-tether.svg)](https://pypi.org/project/fastcrest-tether/)
 
 ![Tether — pip install + tether doctor + tether --help on Modal A10G with TRT EP active](assets/tether-tweet.gif)
 
@@ -28,16 +28,16 @@ The bootstrap installer detects your platform (Mac / Jetson Orin / NVIDIA GPU /
 **Manual install** if you know what you want:
 
 ```bash
-pip install tether                            # core
-pip install 'tether[serve,gpu,monolithic]'    # GPU production path
-pip install 'tether[serve,onnx]'              # Mac / CPU runtime
+pip install fastcrest-tether                            # core
+pip install 'fastcrest-tether[serve,gpu,monolithic]'    # GPU production path
+pip install 'fastcrest-tether[serve,onnx]'              # Mac / CPU runtime
 ```
 
 Requires Python ≥ 3.10.
 
 ### What's new in v0.11.2 (2026-05-29)
 
-- **`tether connect` works on a clean install** — `requests` is now a core dependency, so `tether connect status` no longer raises `ModuleNotFoundError` on `pip install tether` without extras (it had been an undeclared import that only resolved transitively).
+- **`tether connect` works on a clean install** — `requests` is now a core dependency, so `tether connect status` no longer raises `ModuleNotFoundError` on `pip install fastcrest-tether` without extras (it had been an undeclared import that only resolved transitively).
 - **`--fast-kernels` cleared the formal N=100/task L3 LIBERO parity gate** — on Pi0.5 LIBERO-10 tasks 0-2 (600 episodes), Triton fast kernels scored 91.3% (274/300) vs native ORT 85.3% (256/300) — 6.0pp *ahead* of native, so kill-trigger 3 stays clear and the opt-in Triton runtime stays on.
 - **Hardened monolithic serve/bench path, with external-data ONNX** — dedicated ORT provider-options + tokenizer-loading modules are extracted from the request hot path; ONNX models with external weight data (`.onnx` + `.onnx_data`, required once a graph exceeds the 2 GB protobuf limit) now load in both serve and the weight-fusion export pass.
 - **Cleaner streams** — integration-command errors route to stderr, so `--json` consumers and shell pipelines get a clean stdout.
@@ -65,8 +65,8 @@ Breaking: module renames — `tether.exporters.{pi0,smolvla,gr00t}_exporter` →
 We ship patches frequently — make sure you're on the latest:
 
 ```bash
-pip install --upgrade tether              # pip
-uv add --refresh tether                   # uv (the --refresh flag is required;
+pip install --upgrade fastcrest-tether              # pip
+uv add --refresh fastcrest-tether                   # uv (the --refresh flag is required;
                                               # uv caches the package index aggressively
                                               # and won't see new releases without it)
 ```
@@ -83,7 +83,7 @@ only needed for caches built by v0.5.3 or earlier.
 
 ## Performance
 
-`tether[serve,gpu]` uses ONNX Runtime's TensorRT execution provider out of the box. Measured on Modal A10G (Ampere, sm_8.6) on 2026-04-29 against SmolVLA monolithic (5 warmup + 20 measured forward passes, batch=1):
+`fastcrest-tether[serve,gpu]` uses ONNX Runtime's TensorRT execution provider out of the box. Measured on Modal A10G (Ampere, sm_8.6) on 2026-04-29 against SmolVLA monolithic (5 warmup + 20 measured forward passes, batch=1):
 
 | Provider | Mean latency | p95 |
 |---|---|---|
@@ -125,7 +125,7 @@ Full reproducer + 9-iteration debug log: [`reflex_context/03_experiments/2026-04
 Adds ~2 GB to `[serve,gpu]` install (the `tensorrt` package + bundled libs). If you don't want it:
 
 ```bash
-pip install 'tether[serve,gpu-min]'   # ORT-CUDA only, ~5x slower on transformers
+pip install 'fastcrest-tether[serve,gpu-min]'   # ORT-CUDA only, ~5x slower on transformers
 ```
 
 Or disable the `LD_LIBRARY_PATH` patch (e.g. if it conflicts with another env-aware tool):
@@ -275,7 +275,7 @@ Hidden legacy commands (`export`, `bench`, `replay`, etc.) stay callable as alia
 ### Install notes
 
 - `[monolithic]` extra is required for the cos=+1.000000 verified export path (pins transformers==5.3.0)
-- CPU-only: `pip install 'tether[serve,onnx,monolithic]'`
+- CPU-only: `pip install 'fastcrest-tether[serve,onnx,monolithic]'`
 - GPU install needs the FULL cuDNN 9 system library (not just the pip wheel). Easiest path: NVIDIA's container `docker run --gpus all -it nvcr.io/nvidia/tensorrt:24.10-py3`, then `apt-get install -y clang` (for lerobot→evdev), then the pip install
 - `tether serve` errors loudly if cuDNN can't load — no silent CPU fallback
 - First `tether go` downloads weights (~1-14 GB depending on model) — cached on subsequent runs

contrib/ros2/README.md

@@ -1,7 +1,7 @@
 # ROS2 Robot Adapter Starter Kits
 
 **Fork-and-customize** adapters for running Reflex VLA on real robots via ROS2.
-These are NOT part of the core `pip install tether` — they live in `contrib/`
+These are NOT part of the core `pip install fastcrest-tether` — they live in `contrib/`
 and you're expected to adapt them to your specific robot setup.
 
 ## Available Adapters

docs/cli_reference.md

@@ -58,7 +58,7 @@ tether go --model pi05-libero --dry-run
 | `--api-key` | _(none)_ | If set, `/act` requires `X-Tether-Key` header (or `Authorization: Bearer`) |
 | `--dry-run` | `false` | Probe + resolve + print plan; do not pull or serve |
 
-Full flag list: `tether go --help`. Note: models that ship as raw PyTorch require the `[monolithic]` extra (`pip install 'tether[monolithic]'`) for the inline export step.
+Full flag list: `tether go --help`. Note: models that ship as raw PyTorch require the `[monolithic]` extra (`pip install 'fastcrest-tether[monolithic]'`) for the inline export step.
 
 ---
 

docs/eval.md

@@ -62,7 +62,7 @@ Bounded `failure_mode` enum, surfaced in the CLI + telemetry:
 | `egl-black-frames` | Force `MUJOCO_GL=osmesa` in your env. |
 | `dep-version-conflict` | Pin `robosuite==1.4.1`, `bddl==1.0.1`, `mujoco==3.3.2`. Use `--runtime modal` for known-good pins. |
 | `osmesa-compile-hang` | Increase `--preflight-timeout` (cold containers take 60-180s for first-scene compile). |
-| `import-error` | `pip install 'tether[eval-local]'` for local; `--runtime modal` for the bundled image. |
+| `import-error` | `pip install 'fastcrest-tether[eval-local]'` for local; `--runtime modal` for the bundled image. |
 
 The 5th failure (per-episode OOM) is per-call probabilistic; backoff + a legible error in the runner covers it.
 
@@ -145,7 +145,7 @@ Above-`$50` estimate triggers an extra **"are you sure?"** warning so the custom
 Phase 1: **Linux x86_64 only**. Requires the `[eval-local]` extra:
 
 ```bash
-pip install 'tether[eval-local]'
+pip install 'fastcrest-tether[eval-local]'
 ```
 
 `--runtime local` **never silently falls back to Modal**. If the local env is broken, `tether eval` fails loud with a remediation pointer. This avoids surprise Modal bills + masks real env-config issues.

docs/getting_started.md

@@ -1,6 +1,6 @@
 # Getting started with Tether
 
-A 30-minute walkthrough of the typical first hour after `pip install tether[serve,gpu]`.
+A 30-minute walkthrough of the typical first hour after `pip install fastcrest-tether[serve,gpu]`.
 
 This guide assumes a Linux box with an NVIDIA GPU. CPU-only deployments work with `[serve]` instead of `[serve,gpu]` — every example below applies, just replace `--device cuda` with `--device cpu`.
 
@@ -10,10 +10,10 @@ This guide assumes a Linux box with an NVIDIA GPU. CPU-only deployments work wit
 
 ```bash
 # GPU box (requires CUDA 12 + cuDNN 9 — easiest via nvcr.io/nvidia/tensorrt container):
-pip install 'tether[serve,gpu] @ git+https://github.com/FastCrest/tether'
+pip install 'fastcrest-tether[serve,gpu] @ git+https://github.com/FastCrest/tether'
 
 # CPU-only box:
-pip install 'tether[serve,onnx] @ git+https://github.com/FastCrest/tether'
+pip install 'fastcrest-tether[serve,onnx] @ git+https://github.com/FastCrest/tether'
 ```
 
 Or for development from source:
@@ -205,7 +205,7 @@ tether export lerobot/smolvla_base --target orin-nano --output ./sv
 scp -r ./sv jetson:~/sv
 
 # On the Jetson (Jetpack 6.x with TensorRT preinstalled):
-pip install 'tether[serve] @ git+https://github.com/FastCrest/tether'
+pip install 'fastcrest-tether[serve] @ git+https://github.com/FastCrest/tether'
 tether serve ./sv --port 8000 --device cuda
 ```
 
@@ -236,7 +236,7 @@ ORT 1.20+ requires CUDA 12.x + cuDNN 9.x. The pip-installed `nvidia-cudnn-cu12`
 ```bash
 docker run --gpus all -it --rm nvcr.io/nvidia/tensorrt:24.10-py3
 # inside the container:
-pip install 'tether[serve,gpu] @ git+https://github.com/FastCrest/tether'
+pip install 'fastcrest-tether[serve,gpu] @ git+https://github.com/FastCrest/tether'
 tether serve ...
 ```
 

docs/mcp.md

@@ -5,7 +5,7 @@ Tether exposes a [Model Context Protocol](https://spec.modelcontextprotocol.io/)
 ## Install
 
 ```bash
-pip install tether[mcp]
+pip install fastcrest-tether[mcp]
 ```
 
 Pulls [`fastmcp`](https://github.com/jlowin/fastmcp) >= 3.0 alongside the core dependencies.
@@ -89,7 +89,7 @@ Shadow actions, A/B policy routing, and dataset validation run via explicit tool
 
 **"fastmcp not installed"**
 ```bash
-pip install tether[mcp]
+pip install fastcrest-tether[mcp]
 ```
 
 **Claude Desktop doesn't list Tether as a tool**

docs/otel.md

@@ -5,7 +5,7 @@
 ## Install
 
 ```bash
-pip install 'tether[tracing]'
+pip install 'fastcrest-tether[tracing]'
 ```
 
 Without the `[tracing]` extra, tracing no-ops silently; the server emits nothing and costs nothing. Your serve behavior is unchanged.
@@ -102,8 +102,8 @@ File a GitHub issue if either is blocking an integration.
 
 ## Troubleshooting
 
-**"Tracing skipped — pip install tether[tracing] to enable"**
-The `[tracing]` extra isn't installed. Run `pip install 'tether[tracing]'` and restart serve.
+**"Tracing skipped — pip install fastcrest-tether[tracing] to enable"**
+The `[tracing]` extra isn't installed. Run `pip install 'fastcrest-tether[tracing]'` and restart serve.
 
 **Spans appear in Phoenix but not in Datadog**
 Your Datadog Agent probably has `receivers.otlp` disabled. Enable OTLP gRPC on :4317 in `datadog.yaml`, or point `--otel-endpoint` at an OTel Collector that forwards to Datadog.

docs/transports.md

@@ -8,8 +8,8 @@ produces actions; the transport delivers them to the robot client.
 
 | Transport | Flag | When to use | Install |
 |---|---|---|---|
-| **HTTP** (default) | `--transport http` | Standard REST API. Works with any HTTP client (curl, Python requests, browser). Best for prototyping + debugging. | `pip install tether[serve]` |
-| **ZMQ** | `--transport zmq` | Low-latency binary wire. 20× lower bandwidth for multi-camera setups. 10× smaller robot-side install. Best for production robot deployments where every millisecond matters. | Server: `pip install tether[serve]`. Robot: `pip install pyzmq msgpack numpy opencv-python-headless` (~25 MB) |
+| **HTTP** (default) | `--transport http` | Standard REST API. Works with any HTTP client (curl, Python requests, browser). Best for prototyping + debugging. | `pip install fastcrest-tether[serve]` |
+| **ZMQ** | `--transport zmq` | Low-latency binary wire. 20× lower bandwidth for multi-camera setups. 10× smaller robot-side install. Best for production robot deployments where every millisecond matters. | Server: `pip install fastcrest-tether[serve]`. Robot: `pip install pyzmq msgpack numpy opencv-python-headless` (~25 MB) |
 | **ROS2** | (v1.0) | Native ROS2 action server. Reserved for v1.0. | — |
 
 ## Quick Start

docs/troubleshooting.md

@@ -183,7 +183,7 @@ ERROR: Ignored the following versions that require a different python version: 0
 
 **Fix:** On Jetson, install `[serve]` only — **not** `[monolithic]`:
 ```bash
-pip install 'tether[serve]'
+pip install 'fastcrest-tether[serve]'
 ```
 
 The monolithic ONNX export (`tether export --monolithic`) requires lerobot and must run on a **Python 3.12+ host** (desktop, cloud GPU, or Docker). Export there, then copy the ONNX to the Jetson and serve it:

examples/01-chat-quickstart.md

@@ -7,7 +7,7 @@
 ## Install
 
 ```bash
-pip install tether
+pip install fastcrest-tether
 ```
 
 (~40 seconds; pulls torch + transformers + onnx as deps. Yes, it's chunky — the chat agent itself is ~50 KB, but it shares the install with the rest of Tether.)
@@ -55,9 +55,9 @@ Here's what the doctor found:
   ✓ tether 0.3.5
   ⚠ torch + CUDA — torch 2.11.0, CUDA unavailable (you're on Apple Silicon)
   ⚠ ONNX Runtime — not installed
-    Action: pip install 'tether[serve,onnx]'
+    Action: pip install 'fastcrest-tether[serve,onnx]'
   ⚠ fastapi + uvicorn — not installed
-    Action: pip install 'tether[serve,onnx]'
+    Action: pip install 'fastcrest-tether[serve,onnx]'
 ```
 
 ### "What models can I deploy?"

examples/01_circle_tap_so100/README.md

@@ -18,7 +18,7 @@ by 0o8o0 (MIT) per ADR `reflex_context/01_decisions/2026-05-06-vendor-auto-soarm
 
 ```bash
 # Install Tether with the so100 + bench-game extras
-pip install 'tether[so100]'           # adds scservo_sdk for the Pi-side driver
+pip install 'fastcrest-tether[so100]'           # adds scservo_sdk for the Pi-side driver
 pip install lerobot==0.5.1                # for the from-scratch ACT training step
 ```
 

examples/02-deploy-smolvla-jetson.md

@@ -34,7 +34,7 @@ pip install onnxruntime-gpu \
   --index-url https://pypi.jetson-ai-lab.io/jp6/cu126
 
 # 3. Install tether with [serve] only (NOT [gpu], NOT [monolithic])
-pip install 'tether[serve]'
+pip install 'fastcrest-tether[serve]'
 ```
 
 > **Why not `[monolithic]`?** The monolithic export extra depends on `lerobot==0.5.1`, which requires **Python ≥ 3.12**. JetPack 6 ships Python 3.10. Export your model on a desktop/cloud machine with Python 3.12+, then copy the ONNX to the Jetson and serve it.
@@ -49,7 +49,7 @@ echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc
 ### What each piece provides:
 - `numpy<2` — ABI compatibility with Jetson AI Lab pre-built wheels
 - `torch` / `onnxruntime-gpu` (from Jetson AI Lab) — GPU-accelerated inference, compiled for `aarch64` + JetPack CUDA
-- `tether[serve]` — FastAPI + uvicorn HTTP inference server + embodiment validation
+- `fastcrest-tether[serve]` — FastAPI + uvicorn HTTP inference server + embodiment validation
 
 This pulls ~2 GB of dependencies. Takes 5-10 minutes on the Jetson.
 
@@ -61,7 +61,7 @@ Since monolithic export requires Python 3.12+ (for `lerobot`), the typical Jetso
 
 ```bash
 # On your desktop / cloud GPU (Python 3.12+)
-pip install 'tether[serve,monolithic]'
+pip install 'fastcrest-tether[serve,monolithic]'
 tether export --model smolvla-base --out ./smolvla-export/
 ```
 

examples/03-distill-pi05.md

@@ -13,7 +13,7 @@ SnapFlow ([arxiv 2604.05656](https://arxiv.org/abs/2604.05656)) is the canonical
 ## Install
 
 ```bash
-pip install 'tether[monolithic]'
+pip install 'fastcrest-tether[monolithic]'
 ```
 
 ## The command

examples/README.md

@@ -12,7 +12,7 @@ Each example is a self-contained walkthrough you can paste into a terminal. Star
 All examples assume you've installed at least the base package:
 
 ```bash
-pip install tether
+pip install fastcrest-tether
 ```
 
 Some examples need the GPU or monolithic export extras — each example flags what it needs at the top.

examples/so_arm100_smolvla.py

@@ -19,8 +19,8 @@
     - USB-to-serial bridge on /dev/ttyUSB0 (Linux) or /dev/tty.usbserial-* (Mac)
 
 Python requirements:
-    pip install 'tether[serve,gpu,monolithic,lerobot]'   # GPU host
-    pip install 'tether[serve,onnx,lerobot,so100]'       # Mac / Pi at the arm
+    pip install 'fastcrest-tether[serve,gpu,monolithic,lerobot]'   # GPU host
+    pip install 'fastcrest-tether[serve,onnx,lerobot,so100]'       # Mac / Pi at the arm
 
 Calibration:
     If you already have a LeRobot calibration file (recorded via

infra/license-worker/README.md

@@ -90,7 +90,7 @@ python -m tether.admin.issue_license \
 
 ```bash
 # On the customer's machine
-pip install --upgrade tether
+pip install --upgrade fastcrest-tether
 tether pro activate REFLEX-XXXX-XXXX-XXXX
 # ✓ License fetched, signature verified, written to ~/.reflex/pro.license
 # ✓ Hardware bound

infra/telemetry-worker/worker.js

@@ -1,7 +1,7 @@
 /**
  * Reflex telemetry endpoint — Cloudflare Worker.
  *
- * Receives heartbeat POSTs from `pip install tether` deployments
+ * Receives heartbeat POSTs from `pip install fastcrest-tether` deployments
  * running with a valid Pro license OR free-tier telemetry enabled.
  * Validates the payload shape, writes one row per heartbeat to D1,
  * and returns 204 No Content.