voltagegpu-cli 1.3.0__tar.gz → 2.0.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/CHANGELOG.md +153 -87
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/PKG-INFO +111 -10
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/README.md +284 -183
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/pyproject.toml +90 -89
- voltagegpu_cli-2.0.0/tests/fixtures/ark-milan.pem +37 -0
- voltagegpu_cli-2.0.0/tests/fixtures/bundle.json +822 -0
- voltagegpu_cli-2.0.0/tests/fixtures/manifest.json +11 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/tests/test_cli.py +261 -243
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/tests/test_cli_account.py +16 -1
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/tests/test_config.py +11 -0
- voltagegpu_cli-2.0.0/tests/test_hardening.py +165 -0
- voltagegpu_cli-2.0.0/tests/test_vgpu_attest.py +252 -0
- voltagegpu_cli-2.0.0/tests/test_vgpu_cli.py +465 -0
- voltagegpu_cli-2.0.0/tests/test_vgpu_mcp.py +205 -0
- voltagegpu_cli-2.0.0/tests/test_vgpu_remote.py +65 -0
- voltagegpu_cli-2.0.0/tests/test_vgpu_ttl.py +61 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/__init__.py +5 -5
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/cli/cmd_account.py +179 -176
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/cli/cmd_deploy.py +50 -24
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/cli/cmd_misc.py +20 -6
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/cli/cmd_pods.py +24 -15
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/cli/main.py +12 -6
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/cli/output.py +29 -6
- voltagegpu_cli-2.0.0/volt/cli/vgpu.py +475 -0
- voltagegpu_cli-2.0.0/volt/mcp/__init__.py +1 -0
- voltagegpu_cli-2.0.0/volt/mcp/server.py +344 -0
- voltagegpu_cli-2.0.0/volt/sdk/attest.py +296 -0
- voltagegpu_cli-2.0.0/volt/sdk/catalog.py +163 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/sdk/client.py +240 -234
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/sdk/config.py +49 -32
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/sdk/errors.py +2 -1
- voltagegpu_cli-2.0.0/volt/sdk/fleet.py +220 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/sdk/models.py +336 -320
- voltagegpu_cli-2.0.0/volt/sdk/remote.py +129 -0
- voltagegpu_cli-2.0.0/volt/sdk/scripts/snp-proof-in-the-vm.sh +216 -0
- voltagegpu_cli-2.0.0/volt/sdk/scripts/two-proofs-in-the-vm.sh +51 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/sdk/ssh.py +157 -112
- voltagegpu_cli-2.0.0/volt/sdk/ttl.py +104 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/.gitignore +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/LICENSE +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/examples/deploy_standard_pod.py +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/tests/__init__.py +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/tests/conftest.py +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/tests/test_client.py +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/tests/test_live.py +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/tests/test_models.py +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/tests/test_package.py +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/tests/test_ssh.py +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/__main__.py +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/cli/__init__.py +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/cli/cmd_machines.py +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/cli/state.py +0 -0
- {voltagegpu_cli-1.3.0 → voltagegpu_cli-2.0.0}/volt/sdk/__init__.py +0 -0
|
@@ -1,87 +1,153 @@
|
|
|
1
|
-
# Changelog
|
|
2
|
-
|
|
3
|
-
All notable changes to `voltagegpu-cli` are documented here. The format follows
|
|
4
|
-
[Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and the project uses
|
|
5
|
-
[Semantic Versioning](https://semver.org/).
|
|
6
|
-
|
|
7
|
-
## [
|
|
8
|
-
|
|
9
|
-
|
|
10
|
-
|
|
11
|
-
|
|
12
|
-
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
- `
|
|
17
|
-
|
|
18
|
-
-
|
|
19
|
-
(
|
|
20
|
-
|
|
21
|
-
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
- `
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
|
|
30
|
-
|
|
31
|
-
- `voltagegpu
|
|
32
|
-
|
|
33
|
-
- `VOLTAGEGPU_API_KEY` is
|
|
34
|
-
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
39
|
-
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
|
|
43
|
-
|
|
44
|
-
|
|
45
|
-
|
|
46
|
-
|
|
47
|
-
-
|
|
48
|
-
|
|
49
|
-
|
|
50
|
-
|
|
51
|
-
|
|
52
|
-
|
|
53
|
-
|
|
54
|
-
|
|
55
|
-
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
|
|
60
|
-
|
|
61
|
-
|
|
62
|
-
|
|
63
|
-
|
|
64
|
-
|
|
65
|
-
|
|
66
|
-
|
|
67
|
-
|
|
68
|
-
|
|
69
|
-
|
|
70
|
-
|
|
71
|
-
|
|
72
|
-
|
|
73
|
-
|
|
74
|
-
|
|
75
|
-
|
|
76
|
-
|
|
77
|
-
|
|
78
|
-
|
|
79
|
-
|
|
80
|
-
|
|
81
|
-
|
|
82
|
-
- `
|
|
83
|
-
|
|
84
|
-
|
|
85
|
-
|
|
86
|
-
|
|
87
|
-
|
|
1
|
+
# Changelog
|
|
2
|
+
|
|
3
|
+
All notable changes to `voltagegpu-cli` are documented here. The format follows
|
|
4
|
+
[Keep a Changelog](https://keepachangelog.com/en/1.1.0/) and the project uses
|
|
5
|
+
[Semantic Versioning](https://semver.org/).
|
|
6
|
+
|
|
7
|
+
## [2.0.0] - 2026-10-27
|
|
8
|
+
|
|
9
|
+
### Added
|
|
10
|
+
- `vgpu`, a second command for scripts and AI agents: `ls`, `up`, `ps`,
|
|
11
|
+
`exec` (with `-d`), `scp`, `rm`, `attest`, `whoami`. One catalogue for
|
|
12
|
+
Confidential VMs (Intel TDX), Confidential CPU (AMD SEV-SNP), standard GPU
|
|
13
|
+
pods, dedicated machines and sandboxes, cheapest first.
|
|
14
|
+
- `--ttl` on every `vgpu up`, translated to the auto-stop of each product and
|
|
15
|
+
checked against its bounds before anything is sent.
|
|
16
|
+
- `vgpu attest`: challenge drawn locally, embedded proof scripts pinned by
|
|
17
|
+
SHA-256, archive extracted safely and verified locally. For Confidential
|
|
18
|
+
CPU the SEV-SNP report is requested at `--vmpl` (0 by default), and its VMPL
|
|
19
|
+
field must be the level requested (VMPL 1 is no longer required: it was
|
|
20
|
+
snpguest's default request, not the guest's level).
|
|
21
|
+
- `vgpu mcp`, a local MCP server over stdio with the same tools and guards
|
|
22
|
+
(`up` needs `ttl` and `max_hourly_price`, `rm` needs `confirm`), plus
|
|
23
|
+
`inference` on the OpenAI-compatible API with a `vgpu_` key.
|
|
24
|
+
- Documented exit codes, including 6 for HTTP 402 (`API_KEY_BUDGET_EXCEEDED`)
|
|
25
|
+
and 4 for `API_KEY_SCOPE_MISSING`.
|
|
26
|
+
- `vgpu up --ssh-key-id` (`ssh_key_ids` in `vgpu mcp`) on a standard GPU too:
|
|
27
|
+
one key id, needed when the account has several SSH keys. Without it, the
|
|
28
|
+
API's 400 becomes a clear message (exit 2, `SSH_KEY_ID_REQUIRED`).
|
|
29
|
+
|
|
30
|
+
### Changed
|
|
31
|
+
- `voltagegpu balance` no longer shows `$0.00` when the key cannot read the
|
|
32
|
+
balance (agent key without the billing scope): it says the balance is hidden.
|
|
33
|
+
- A `vgpu_` inference key in `VOLTAGEGPU_API_KEY` is no longer taken for the
|
|
34
|
+
pod API key; the next source (config file) is used.
|
|
35
|
+
- `voltagegpu` and `volt` are otherwise unchanged. The major version marks the new
|
|
36
|
+
command and its exit-code contract.
|
|
37
|
+
|
|
38
|
+
## [1.3.1] - 2026-10-10
|
|
39
|
+
|
|
40
|
+
Security hardening after an independent review of 1.3.0. Upgrade recommended.
|
|
41
|
+
|
|
42
|
+
### Security
|
|
43
|
+
- The SSH command returned by the API is only ever displayed in its validated,
|
|
44
|
+
rebuilt form (`ssh -p PORT user@host`). A command of any other form is
|
|
45
|
+
neither shown nor run, so a tampered answer cannot end up copied into a
|
|
46
|
+
shell from `pods show`, `pods wait` or `deploy --wait`.
|
|
47
|
+
- Server text (machine names, locations, error messages, ids) can no longer
|
|
48
|
+
inject terminal escape sequences or rich markup: control characters are
|
|
49
|
+
stripped and the text is printed literally. Before, an odd string could also
|
|
50
|
+
crash the CLI right after a deploy.
|
|
51
|
+
- `voltagegpu verify` calls voltage-verify in-process instead of
|
|
52
|
+
`python -m voltage_verify`, which put the current directory first on the
|
|
53
|
+
module path (a `voltage_verify.py` next to a downloaded bundle would run).
|
|
54
|
+
- `voltagegpu ssh` looks for `ssh` on PATH only, never in the current
|
|
55
|
+
directory (Windows `shutil.which` checks it first), and never runs a
|
|
56
|
+
`.bat`/`.cmd` script.
|
|
57
|
+
- The one-time sudo password of a Confidential VM is printed exactly as
|
|
58
|
+
received (no markup or emoji parsing could alter it).
|
|
59
|
+
- The API key is only sent to `voltagegpu.com` hosts (or localhost) unless
|
|
60
|
+
`VOLT_ALLOW_CUSTOM_BASE_URL=1` is set.
|
|
61
|
+
- A malformed `~/.volt/config.ini` no longer prints its content (which may be
|
|
62
|
+
the key) in an error; an existing file is set to 0600 before the key is
|
|
63
|
+
written.
|
|
64
|
+
|
|
65
|
+
### Changed
|
|
66
|
+
- `deploy standard --cheapest --yes` requires `--max-price`, so a script never
|
|
67
|
+
rents an expensive machine by surprise.
|
|
68
|
+
- If the deploy request gets no answer, the CLI says the deploy may have gone
|
|
69
|
+
through and to check `voltagegpu pods list`, instead of suggesting a retry.
|
|
70
|
+
- A warning is printed if the hourly price rose between the listing and the
|
|
71
|
+
deploy.
|
|
72
|
+
|
|
73
|
+
## [1.3.0] - 2026-10-10
|
|
74
|
+
|
|
75
|
+
Realigns the CLI with the product as it is today: Confidential VMs (Intel TDX)
|
|
76
|
+
and standard pods, both deployable from the public API. Every command was run
|
|
77
|
+
against the production API before release, including a paid deploy, SSH and
|
|
78
|
+
release of a standard pod.
|
|
79
|
+
|
|
80
|
+
### Added
|
|
81
|
+
- `voltagegpu` command (the `volt` command still works, same program).
|
|
82
|
+
- `voltagegpu machines [--standard | --confidential] [--gpu] [--max-price] [--min-gpus] [--json]`:
|
|
83
|
+
standard machines and Confidential VM tiers with the hourly price charged.
|
|
84
|
+
- `voltagegpu deploy standard --machine ID | --cheapest`: deploy a standard pod
|
|
85
|
+
(no enclave), with `--template`, `--ssh-key`, `--jupyter`, `--ports`,
|
|
86
|
+
`--auto-terminate`, `--wait`. Asks for confirmation before charging; scripts
|
|
87
|
+
pass `--yes`.
|
|
88
|
+
- `voltagegpu deploy confidential --tier TIER`: deploy a Confidential VM, with
|
|
89
|
+
`--hardstop`, repeatable `--port 8000[/tcp|udp]` and `--wait`. Prints the
|
|
90
|
+
one-time sudo password and the in-VM attestation commands.
|
|
91
|
+
- `voltagegpu templates`: images available for standard pods.
|
|
92
|
+
- `voltagegpu pods wait POD_ID`: wait until SSH is ready.
|
|
93
|
+
- `voltagegpu ssh POD_ID [-- COMMAND]`: run ssh directly (or `--print` it).
|
|
94
|
+
- `voltagegpu balance`: balance and running pods (`GET /api/volt/balance`).
|
|
95
|
+
- `voltagegpu login` / `logout`: store the API key in `~/.volt/config.ini`,
|
|
96
|
+
created with mode 0600, after checking that the API accepts it.
|
|
97
|
+
- `voltagegpu verify ...`: verifies a Confidential VM attestation bundle by
|
|
98
|
+
running voltage-verify (`pip install "voltagegpu-cli[verify]"`).
|
|
99
|
+
- `VOLTAGEGPU_API_KEY` is accepted as an alias of `VOLT_API_KEY`.
|
|
100
|
+
- SDK: `list_machines`, `list_templates`, `deploy_standard`,
|
|
101
|
+
`deploy_confidential`, `release_pod`, `wait_until_reachable`, `get_balance`;
|
|
102
|
+
typed immutable models; `parse_ssh_command`.
|
|
103
|
+
- Test suite (unit, CLI and opt-in live checks) and a CI workflow.
|
|
104
|
+
|
|
105
|
+
### Changed
|
|
106
|
+
- `pods stop` now releases the pod (`DELETE /api/volt/pods/:id`) and prints
|
|
107
|
+
the refunded unused prepaid time. Standard pods cannot be paused, so the old
|
|
108
|
+
`POST .../stop` call failed for them. `pods delete` and `pods get` remain as
|
|
109
|
+
aliases of `pods stop` and `pods show`.
|
|
110
|
+
- Errors show the API's own message (for example "Insufficient balance. One
|
|
111
|
+
hour costs $0.59...") instead of a raw HTTP exception, and never a traceback
|
|
112
|
+
unless `VOLT_DEBUG=1`.
|
|
113
|
+
- Python 3.10 or newer is required (3.9 reached end of life in October 2025).
|
|
114
|
+
|
|
115
|
+
### Removed
|
|
116
|
+
- `volt cc inventory`, `volt cc deploy` and `volt templates get`: Confidential
|
|
117
|
+
containers were retired on 2026-10-01. The commands now print how to deploy a
|
|
118
|
+
Confidential VM instead.
|
|
119
|
+
- `volt account info`, which called an endpoint that does not exist.
|
|
120
|
+
|
|
121
|
+
### Fixed
|
|
122
|
+
- `ssh-keys add` sent `publicKey`; the API expects `public_key`, so every key
|
|
123
|
+
upload failed with HTTP 400. It also refuses a private key file now.
|
|
124
|
+
- `account balance` called a dashboard-only endpoint and always failed with
|
|
125
|
+
HTTP 401.
|
|
126
|
+
- `cc inventory` called a dashboard-only endpoint and reported "no GPUs
|
|
127
|
+
available" instead of an error.
|
|
128
|
+
- Pod ids are validated before they are put into a URL path.
|
|
129
|
+
|
|
130
|
+
### Security
|
|
131
|
+
- The SSH command returned by the API is parsed into user, host and port, each
|
|
132
|
+
checked against a strict pattern, and ssh is started from a list of
|
|
133
|
+
arguments, never through a shell. A tampered answer such as
|
|
134
|
+
`ssh -o ProxyCommand=...` is refused.
|
|
135
|
+
- Plain `http://` base URLs are refused (except localhost), redirects are never
|
|
136
|
+
followed, and TLS certificates are always verified, so the API key cannot be
|
|
137
|
+
sent in clear or to another host.
|
|
138
|
+
- The API key is never printed (only `volt_****` and its last four
|
|
139
|
+
characters), including in `repr()` of the configuration.
|
|
140
|
+
|
|
141
|
+
## [1.2.1] - 2026-04-11
|
|
142
|
+
|
|
143
|
+
- Confidential container templates (`volt templates`), the machines list and
|
|
144
|
+
`volt cc deploy --template`.
|
|
145
|
+
|
|
146
|
+
## [1.2.0] - 2026-04-11
|
|
147
|
+
|
|
148
|
+
- `volt cc` command group for Confidential containers; default API base URL
|
|
149
|
+
`https://api.voltagegpu.com/api`; JSON output on list commands.
|
|
150
|
+
|
|
151
|
+
## [1.1.0] and [1.0.0]
|
|
152
|
+
|
|
153
|
+
- First releases of the CLI and Python SDK.
|
|
@@ -1,13 +1,13 @@
|
|
|
1
1
|
Metadata-Version: 2.5
|
|
2
2
|
Name: voltagegpu-cli
|
|
3
|
-
Version:
|
|
4
|
-
Summary: VoltageGPU CLI and Python SDK:
|
|
3
|
+
Version: 2.0.0
|
|
4
|
+
Summary: VoltageGPU CLI and Python SDK: Confidential VMs (Intel TDX), Confidential CPU (AMD SEV-SNP), standard GPU pods, dedicated machines and sandboxes, with the vgpu command for scripts and AI agents.
|
|
5
5
|
Project-URL: Homepage, https://voltagegpu.com
|
|
6
6
|
Project-URL: Documentation, https://docs.voltagegpu.com/cli
|
|
7
7
|
Project-URL: API reference, https://docs.voltagegpu.com/api-reference
|
|
8
|
-
Project-URL: Source, https://github.com/
|
|
9
|
-
Project-URL: Changelog, https://github.com/
|
|
10
|
-
Project-URL: Issues, https://github.com/
|
|
8
|
+
Project-URL: Source, https://github.com/VoltageGPU/VOLTAGEGPU-CLI/tree/release/1.3
|
|
9
|
+
Project-URL: Changelog, https://github.com/VoltageGPU/VOLTAGEGPU-CLI/blob/release/1.3/CHANGELOG.md
|
|
10
|
+
Project-URL: Issues, https://github.com/VoltageGPU/VOLTAGEGPU-CLI/issues
|
|
11
11
|
Author-email: "VoltageGPU (VOLTAGE EI)" <contact@voltagegpu.com>
|
|
12
12
|
License: MIT License
|
|
13
13
|
|
|
@@ -169,7 +169,7 @@ Every list command takes `--json`; deploys and stops take `--yes`.
|
|
|
169
169
|
|
|
170
170
|
```bash
|
|
171
171
|
# Deploy the cheapest standard machine, wait for SSH, run a job, release it.
|
|
172
|
-
POD=$(voltagegpu deploy standard --cheapest --yes --json | jq -r .deploy.pod_id)
|
|
172
|
+
POD=$(voltagegpu deploy standard --cheapest --max-price 1 --yes --json | jq -r .deploy.pod_id)
|
|
173
173
|
voltagegpu pods wait "$POD"
|
|
174
174
|
voltagegpu ssh "$POD" -o StrictHostKeyChecking=accept-new -- nvidia-smi
|
|
175
175
|
voltagegpu pods stop "$POD" --yes
|
|
@@ -178,6 +178,106 @@ voltagegpu pods stop "$POD" --yes
|
|
|
178
178
|
Exit codes: `0` success, `1` API or runtime error, `2` usage error or missing
|
|
179
179
|
confirmation.
|
|
180
180
|
|
|
181
|
+
## `vgpu`: the short command for scripts and AI agents
|
|
182
|
+
|
|
183
|
+
Installed with the package, next to `voltagegpu`. The key comes from
|
|
184
|
+
`VOLT_API_KEY` or `VOLTAGEGPU_API_KEY` (or `voltagegpu login`) and is never
|
|
185
|
+
printed. Give an agent a limited key (scopes, its own machines only, a
|
|
186
|
+
daily budget) from https://voltagegpu.com/api-keys, never your full key.
|
|
187
|
+
|
|
188
|
+
```bash
|
|
189
|
+
vgpu ls --json # every offer, cheapest first
|
|
190
|
+
vgpu ls --kind vm --gpu H100 # vm | cpu | standard | baremetal | sandbox
|
|
191
|
+
vgpu up rtx6000b-small --name train1 --ttl 2h -y --json
|
|
192
|
+
vgpu ps --json
|
|
193
|
+
vgpu exec train1 "nvidia-smi" # returns the remote exit code
|
|
194
|
+
vgpu exec train1 -d "python train.py" # background: prints PID and log path
|
|
195
|
+
vgpu scp train1 /home/ubuntu/out.tgz ./out.tgz # --up to copy the other way
|
|
196
|
+
vgpu attest train1 # Confidential VM or Confidential CPU only
|
|
197
|
+
vgpu rm train1 -y
|
|
198
|
+
vgpu whoami # scopes, budgets, spend, balance
|
|
199
|
+
vgpu mcp # local MCP server (stdio) for agents
|
|
200
|
+
```
|
|
201
|
+
|
|
202
|
+
`--ttl` is mandatory and becomes the auto-stop of each product, rounded up,
|
|
203
|
+
refused locally when out of bounds: Confidential VM and CPU up to 720h,
|
|
204
|
+
standard pods 1h to 720h (whole hours), dedicated machines 1h to 72h (whole
|
|
205
|
+
hours), sandboxes up to 7d. `up` asks for confirmation unless `-y` is given,
|
|
206
|
+
and fails without `-y` when not run in a terminal. A deploy that gets no
|
|
207
|
+
answer is never retried: run `vgpu ps` first.
|
|
208
|
+
|
|
209
|
+
`exec`, `scp` and `attest` use your local SSH key (`-i`, or `VGPU_SSH_KEY`).
|
|
210
|
+
The first connection to a machine records its host key in
|
|
211
|
+
`~/.volt/known_hosts.d/<id>` (trust on first use); a changed key is refused.
|
|
212
|
+
Names given at `up` are kept in `~/.volt/vgpu-names.json`; ids work too.
|
|
213
|
+
|
|
214
|
+
Confidential VMs and CPU tiers bind the SSH keys already on the account.
|
|
215
|
+
A standard GPU binds one key: the account's only key, or `--ssh-key-id ID`
|
|
216
|
+
when the account has several (`voltagegpu ssh-keys list` shows the ids).
|
|
217
|
+
Dedicated machines and sandboxes take key ids: by default every key of the
|
|
218
|
+
account, but a key limited to its own machines cannot read them, so pass
|
|
219
|
+
`--ssh-key-id ID` (repeatable). For such a key `whoami` and `voltagegpu
|
|
220
|
+
balance` show the balance as hidden: only a full key reads it.
|
|
221
|
+
|
|
222
|
+
`vgpu attest` draws a challenge on your machine, runs the published proof
|
|
223
|
+
script in the machine (`two-proofs-in-the-vm.sh` for a Confidential VM,
|
|
224
|
+
`snp-proof-in-the-vm.sh` for a Confidential CPU tier, both embedded and pinned
|
|
225
|
+
by SHA-256), copies the archive back and verifies it here: `voltage-verify`
|
|
226
|
+
for Intel TDX + NVIDIA (`pip install 'voltagegpu-cli[verify]'`, `--offline`
|
|
227
|
+
for no network). For SEV-SNP: the report is requested at `--vmpl` (0 by
|
|
228
|
+
default, sent as `SNP_VMPL`; if VMPL 0 is refused, the script says so and asks
|
|
229
|
+
again at 1, and the result reports it), REPORT_DATA equals the challenge, the
|
|
230
|
+
report policy forbids debug and its VMPL field is the level requested (the
|
|
231
|
+
guest kernel uses VMPCK0, consistent with VMPL 0, no SVSM), then the AMD chain and the VCEK are
|
|
232
|
+
fetched again from AMD into a new folder (`snpguest 0.11.0 fetch ca` / `fetch
|
|
233
|
+
vcek` on your machine), the ARK must be one of the AMD roots pinned in the
|
|
234
|
+
script, and `snpguest verify certs` / `verify attestation` run with those
|
|
235
|
+
certificates, never the ones in the archive (they come from the VM). Without
|
|
236
|
+
a local `snpguest`, or when the AMD key server cannot be reached, the result
|
|
237
|
+
is `INCOMPLETE`, never `VERIFIED`. Standard pods, dedicated machines and sandboxes have no tenant
|
|
238
|
+
attestation and are refused.
|
|
239
|
+
|
|
240
|
+
| Exit code | Meaning |
|
|
241
|
+
|---|---|
|
|
242
|
+
| 0 | ok |
|
|
243
|
+
| 1 | error (including a deploy that got no answer) |
|
|
244
|
+
| 2 | usage: bad option, `--ttl` out of bounds, missing `-y` |
|
|
245
|
+
| 3 | API key missing, malformed or refused |
|
|
246
|
+
| 4 | API key scope missing (`API_KEY_SCOPE_MISSING`), or account resource refused to a key limited to its own machines (`API_KEY_ACCOUNT_RESOURCE`) |
|
|
247
|
+
| 5 | not found (machine or tier) |
|
|
248
|
+
| 6 | balance too low or API key budget exceeded (`API_KEY_BUDGET_EXCEEDED`, HTTP 402): stop, nothing was charged |
|
|
249
|
+
| 7 | unavailable: sold out, over `--max-price`, rate limited, offer closed |
|
|
250
|
+
| 8 | attestation failed |
|
|
251
|
+
| 9 | attestation incomplete, or a local tool is missing (ssh, scp, voltage-verify) |
|
|
252
|
+
|
|
253
|
+
With `--json`, an error is also printed on stdout as
|
|
254
|
+
`{"error": {"code", "message", "exit_code", ...}}`.
|
|
255
|
+
|
|
256
|
+
### Local MCP server
|
|
257
|
+
|
|
258
|
+
`vgpu mcp` serves the tools `ls`, `up`, `ps`, `exec`, `scp`, `rm`, `attest`,
|
|
259
|
+
`inference` and `whoami` over stdio. It runs on your machine because `exec`,
|
|
260
|
+
`scp` and `attest` use your SSH private key, which never leaves it. Same
|
|
261
|
+
guards as the command line: `up` requires `ttl` and `max_hourly_price`, `rm`
|
|
262
|
+
requires `confirm: true`, a 402 comes back as `API_KEY_BUDGET_EXCEEDED` with
|
|
263
|
+
"do not retry", the key is never part of an answer, and the sudo password of
|
|
264
|
+
a Confidential VM is never handed to the agent. `inference` calls the
|
|
265
|
+
OpenAI-compatible API with an inference key (`vgpu_...`) in
|
|
266
|
+
`VOLTAGEGPU_API_KEY`; a `vgpu_` value there is never sent to the pod API,
|
|
267
|
+
which takes the `volt_` key from `VOLT_API_KEY`.
|
|
268
|
+
|
|
269
|
+
```bash
|
|
270
|
+
claude mcp add voltagegpu -e VOLT_API_KEY=volt_... -- vgpu mcp # Claude Code
|
|
271
|
+
```
|
|
272
|
+
|
|
273
|
+
```toml
|
|
274
|
+
# Codex: ~/.codex/config.toml
|
|
275
|
+
[mcp_servers.voltagegpu]
|
|
276
|
+
command = "vgpu"
|
|
277
|
+
args = ["mcp"]
|
|
278
|
+
env = { VOLT_API_KEY = "volt_..." }
|
|
279
|
+
```
|
|
280
|
+
|
|
181
281
|
## Python SDK
|
|
182
282
|
|
|
183
283
|
```python
|
|
@@ -211,7 +311,7 @@ curl https://api.voltagegpu.com/api/volt/pods -H "X-API-Key: $VOLT_API_KEY"
|
|
|
211
311
|
| Setting | Where |
|
|
212
312
|
| --- | --- |
|
|
213
313
|
| API key | `--api-key` of `login`, `VOLT_API_KEY` (or `VOLTAGEGPU_API_KEY`), then `~/.volt/config.ini` |
|
|
214
|
-
| API base URL | `VOLT_BASE_URL` (https only
|
|
314
|
+
| API base URL | `VOLT_BASE_URL` (https only, voltagegpu.com hosts unless `VOLT_ALLOW_CUSTOM_BASE_URL=1`; default `https://api.voltagegpu.com/api`) |
|
|
215
315
|
| Debug | `VOLT_DEBUG=1` shows tracebacks |
|
|
216
316
|
|
|
217
317
|
`voltagegpu config` shows where the key comes from, masked.
|
|
@@ -222,9 +322,10 @@ curl https://api.voltagegpu.com/api/volt/pods -H "X-API-Key: $VOLT_API_KEY"
|
|
|
222
322
|
never followed and certificates are always verified.
|
|
223
323
|
- The key is never printed. `~/.volt/config.ini` is created readable by you
|
|
224
324
|
only, and `voltagegpu config` warns if its permissions are too open.
|
|
225
|
-
- The SSH command returned by the API is parsed into user, host and port
|
|
226
|
-
|
|
227
|
-
|
|
325
|
+
- The SSH command returned by the API is parsed into user, host and port, only
|
|
326
|
+
that rebuilt form is ever shown, and ssh is started without a shell (and never
|
|
327
|
+
from the current directory), so a tampered answer cannot run a command on
|
|
328
|
+
your machine. Server text is printed literally, without terminal escapes.
|
|
228
329
|
- `ssh-keys add` refuses private keys.
|
|
229
330
|
|
|
230
331
|
Report a vulnerability to contact@voltagegpu.com.
|