flashnode 0.3.5__tar.gz → 0.4.1__tar.gz

{flashnode-0.3.5 → flashnode-0.4.1}/PKG-INFO

@@ -1,13 +1,13 @@
 Metadata-Version: 2.4
 Name: flashnode
-Version: 0.3.5
+Version: 0.4.1
 Summary: Open host agent for the FlashML fragmented-compute network: join, benchmark, execute sandboxed ML tasks, earn contribution credits.
 License: Apache-2.0
 Project-URL: Homepage, https://github.com/Zolli-Labs/flashnode
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: flashruntime<0.6,>=0.4.3
+Requires-Dist: flashruntime<0.7,>=0.6.1
 Requires-Dist: psutil>=5.9
 Requires-Dist: websockets>=12
 Requires-Dist: cryptography>=42
@@ -49,6 +49,42 @@ verified results. Two profiles:
   inside managed pools (DaemonSet); KubeRay owns workload pods there.
   Implemented.
 
+## Install
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/Zolli-Labs/flashml/main/install.sh | sh
+```
+
+One line, no root, nothing system-wide. It puts the agent in its own
+environment under `~/.flashnode/agent` and links `flash` (and `flashnode`)
+into `~/.local/bin`. **If this machine has no suitable Python, it downloads
+one** — that is the point: a RunPod pod shipping 3.8 and an Alibaba FC
+sandbox shipping 3.13 both end up running the agent on the same interpreter.
+
+Before the installer is on `main`, take it from the branch and say which
+version to install:
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/Zolli-Labs/flashml/fix/trusted-tier-execution/install.sh \
+  | FLASH_REF="flashnode @ git+https://github.com/Zolli-Labs/flashml.git@fix/trusted-tier-execution#subdirectory=flashnode" sh
+```
+
+Then, wherever you installed it from:
+
+```bash
+flash login dev        # or: prod, or any coordinator URL
+flash run -d           # contribute, in the background
+flash logs             # watch what it is doing
+flash stop             # ask it to finish up and exit
+```
+
+`flash` and `flashnode` are the same program under two names, and every
+command takes an optional SERVER: a nickname (`dev`, `prod`), a URL, or
+enough of a coordinator you have already joined to be unambiguous
+(`flash run render`). Leave it out and — if this machine joined exactly one
+coordinator — that is the one it means. `--coordinator URL` still works
+everywhere it ever did.
+
 ## Check your machine first
 
 ```bash
@@ -119,6 +155,18 @@ flashnode work --coordinator http://<coordinator>:8100   # reads the saved token
 flashnode logout --coordinator http://<coordinator>:8100 # forget it locally (does not revoke server-side)
 ```
 
+FlashML Cloud also supports browser device-code login. On a rented or
+disposable host, add `--ephemeral`:
+
+```bash
+flashnode login --coordinator https://<cloud-api> --ephemeral
+```
+
+That flag rotates the local node id before enrollment, so a reused RunPod or
+Colab state directory cannot carry the previous renter's machine identity into
+the next account. Do not use it on a laptop or workstation whose identity
+should remain stable while offline.
+
 `login`/`logout` write to a per-coordinator credential store at
 `~/.flashnode/credentials.json` (override with `FLASHNODE_CREDENTIALS`),
 keyed by coordinator URL so one machine can hold separate tokens for
@@ -126,11 +174,10 @@ separate pools. The file is written with mode `0600` on every save. A
 missing or unparseable file is treated as "no saved token" rather than a
 crash. `CoordinatorClient` sends the saved token as a bearer header on
 every request to that coordinator once it's saved — there is nothing else
-to configure. Token issuance is still manual and out-of-band today (the
-coordinator operator hands you the token; there is no self-service signup
-or browser device flow yet), and `flashnode logout` only removes the local
-copy — the operator revokes access by removing your token from the
-coordinator's configuration.
+to configure. Self-hosted coordinators may still issue tokens manually;
+FlashML Cloud uses the browser device-code flow. `flashnode logout` only
+removes the local copy — revoke the machine in the cloud console or through
+the coordinator's own access controls as appropriate.
 
 - Stable node identity; registers with capabilities (CPU, RAM, arch, GPU)
   and **re-registers automatically** if the coordinator restarts.

{flashnode-0.3.5 → flashnode-0.4.1}/README.md

@@ -32,6 +32,42 @@ verified results. Two profiles:
   inside managed pools (DaemonSet); KubeRay owns workload pods there.
   Implemented.
 
+## Install
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/Zolli-Labs/flashml/main/install.sh | sh
+```
+
+One line, no root, nothing system-wide. It puts the agent in its own
+environment under `~/.flashnode/agent` and links `flash` (and `flashnode`)
+into `~/.local/bin`. **If this machine has no suitable Python, it downloads
+one** — that is the point: a RunPod pod shipping 3.8 and an Alibaba FC
+sandbox shipping 3.13 both end up running the agent on the same interpreter.
+
+Before the installer is on `main`, take it from the branch and say which
+version to install:
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/Zolli-Labs/flashml/fix/trusted-tier-execution/install.sh \
+  | FLASH_REF="flashnode @ git+https://github.com/Zolli-Labs/flashml.git@fix/trusted-tier-execution#subdirectory=flashnode" sh
+```
+
+Then, wherever you installed it from:
+
+```bash
+flash login dev        # or: prod, or any coordinator URL
+flash run -d           # contribute, in the background
+flash logs             # watch what it is doing
+flash stop             # ask it to finish up and exit
+```
+
+`flash` and `flashnode` are the same program under two names, and every
+command takes an optional SERVER: a nickname (`dev`, `prod`), a URL, or
+enough of a coordinator you have already joined to be unambiguous
+(`flash run render`). Leave it out and — if this machine joined exactly one
+coordinator — that is the one it means. `--coordinator URL` still works
+everywhere it ever did.
+
 ## Check your machine first
 
 ```bash
@@ -102,6 +138,18 @@ flashnode work --coordinator http://<coordinator>:8100   # reads the saved token
 flashnode logout --coordinator http://<coordinator>:8100 # forget it locally (does not revoke server-side)
 ```
 
+FlashML Cloud also supports browser device-code login. On a rented or
+disposable host, add `--ephemeral`:
+
+```bash
+flashnode login --coordinator https://<cloud-api> --ephemeral
+```
+
+That flag rotates the local node id before enrollment, so a reused RunPod or
+Colab state directory cannot carry the previous renter's machine identity into
+the next account. Do not use it on a laptop or workstation whose identity
+should remain stable while offline.
+
 `login`/`logout` write to a per-coordinator credential store at
 `~/.flashnode/credentials.json` (override with `FLASHNODE_CREDENTIALS`),
 keyed by coordinator URL so one machine can hold separate tokens for
@@ -109,11 +157,10 @@ separate pools. The file is written with mode `0600` on every save. A
 missing or unparseable file is treated as "no saved token" rather than a
 crash. `CoordinatorClient` sends the saved token as a bearer header on
 every request to that coordinator once it's saved — there is nothing else
-to configure. Token issuance is still manual and out-of-band today (the
-coordinator operator hands you the token; there is no self-service signup
-or browser device flow yet), and `flashnode logout` only removes the local
-copy — the operator revokes access by removing your token from the
-coordinator's configuration.
+to configure. Self-hosted coordinators may still issue tokens manually;
+FlashML Cloud uses the browser device-code flow. `flashnode logout` only
+removes the local copy — revoke the machine in the cloud console or through
+the coordinator's own access controls as appropriate.
 
 - Stable node identity; registers with capabilities (CPU, RAM, arch, GPU)
   and **re-registers automatically** if the coordinator restarts.