localdeploy 0.6.1__tar.gz → 0.6.2__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.
- {localdeploy-0.6.1 → localdeploy-0.6.2}/CHANGELOG.md +21 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/PKG-INFO +74 -25
- {localdeploy-0.6.1 → localdeploy-0.6.2}/README.md +64 -15
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/__init__.py +1 -1
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/web/index.html +1 -1
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/web/js/benchmark-views.js +2 -2
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/web/js/benchmark.js +3 -3
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/web/js/models.js +1 -1
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/web/styles.css +6 -6
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy.egg-info/PKG-INFO +74 -25
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy.egg-info/requires.txt +8 -8
- {localdeploy-0.6.1 → localdeploy-0.6.2}/pyproject.toml +2 -2
- localdeploy-0.6.2/requirements.txt +9 -0
- localdeploy-0.6.1/requirements.txt +0 -9
- {localdeploy-0.6.1 → localdeploy-0.6.2}/.env.example +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/LICENSE +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/MANIFEST.in +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/SECURITY.md +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/api_server.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/benchmark.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/chat_cli.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/compare_models.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/config.example.json +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/backends/__init__.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/backends/llamacpp.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/backends/ollama.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/backends/openai_compatible.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/benchmark.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/chat_cli.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/cli.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/compare_models.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/__init__.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/_config.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/_ollama.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/bakeoff.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/bench.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/calibration.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/community.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/fit.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/hardware.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/history.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/manifest.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/models.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/monitor.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/profiles.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/recommend.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/registry.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/report.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/starter.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/control/updates.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/grader_sandbox.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/server.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/utils.py +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/web/favicon.png +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/web/js/app.js +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/web/js/chat.js +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/web/js/shared.js +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/web/js/system.js +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy/web/logo.svg +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy.egg-info/SOURCES.txt +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy.egg-info/dependency_links.txt +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy.egg-info/entry_points.txt +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/localdeploy.egg-info/top_level.txt +0 -0
- {localdeploy-0.6.1 → localdeploy-0.6.2}/setup.cfg +0 -0
|
@@ -2,6 +2,27 @@
|
|
|
2
2
|
|
|
3
3
|
All notable changes to this project should be documented here.
|
|
4
4
|
|
|
5
|
+
## 0.6.2 - 2026-10-03
|
|
6
|
+
|
|
7
|
+
- Benchmark screen: the leaderboard and comparison tables said "N/N passed" even for a model with low
|
|
8
|
+
accuracy, because the number counts requests that finished without an error. They now say "completed".
|
|
9
|
+
The `passed` field in saved runs and exports is unchanged.
|
|
10
|
+
- "Contribute anonymized benchmark" is now "Save anonymized benchmark", and its dialog says that the
|
|
11
|
+
snapshot is saved to a file on this computer and nothing is sent anywhere. That was already how it
|
|
12
|
+
worked; the old wording suggested an upload.
|
|
13
|
+
- The category heatmap and the pass/fail matrix used pale text colours made for the dark theme, so the
|
|
14
|
+
category names were nearly invisible in the light theme. They now use the theme's text colour.
|
|
15
|
+
- Launchers: the root launchers (`run.bat`, `run.ps1`, `run.command`, `run.sh`) share one lifecycle,
|
|
16
|
+
report install failures more clearly, and `run.sh` and `run.command` are executable in the repository.
|
|
17
|
+
Install and run documentation was aligned with them.
|
|
18
|
+
- README: opens with the demo and a short quick start, the Docker example uses the `latest` tag, and the
|
|
19
|
+
images use absolute URLs so they render on the PyPI project page.
|
|
20
|
+
- The PyPI summary now matches the repository description.
|
|
21
|
+
- Refreshed the animated product walkthrough.
|
|
22
|
+
- Updated runtime dependencies, including FastAPI 0.141.1, huggingface-hub 1.31.0 and python-dotenv
|
|
23
|
+
1.2.3, and the GitHub Actions used by CI and the publish workflows. CI now caches pip, and Dependabot
|
|
24
|
+
is configured with grouped updates.
|
|
25
|
+
|
|
5
26
|
## 0.6.1 - 2026-08-24
|
|
6
27
|
|
|
7
28
|
- Published the Python package to PyPI and added a GHCR container publishing workflow with a
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: localdeploy
|
|
3
|
-
Version: 0.6.
|
|
4
|
-
Summary:
|
|
3
|
+
Version: 0.6.2
|
|
4
|
+
Summary: Find, deploy, manage, and benchmark local AI models for your hardware across Ollama, llama.cpp, and compatible runtimes.
|
|
5
5
|
Author: oney-erge
|
|
6
6
|
License-Expression: MIT
|
|
7
7
|
Project-URL: Homepage, https://github.com/oney-erge/LocalDeploy
|
|
@@ -23,19 +23,19 @@ Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
|
23
23
|
Requires-Python: >=3.10
|
|
24
24
|
Description-Content-Type: text/markdown
|
|
25
25
|
License-File: LICENSE
|
|
26
|
-
Requires-Dist: fastapi==0.
|
|
27
|
-
Requires-Dist: uvicorn==0.
|
|
28
|
-
Requires-Dist: requests==2.
|
|
29
|
-
Requires-Dist: python-dotenv==1.2.
|
|
30
|
-
Requires-Dist: pydantic==2.13.
|
|
31
|
-
Requires-Dist: python-multipart==0.0.
|
|
32
|
-
Requires-Dist: huggingface_hub==1.
|
|
26
|
+
Requires-Dist: fastapi==0.141.1
|
|
27
|
+
Requires-Dist: uvicorn==0.53.0
|
|
28
|
+
Requires-Dist: requests==2.34.2
|
|
29
|
+
Requires-Dist: python-dotenv==1.2.3
|
|
30
|
+
Requires-Dist: pydantic==2.13.5
|
|
31
|
+
Requires-Dist: python-multipart==0.0.32
|
|
32
|
+
Requires-Dist: huggingface_hub==1.31.0
|
|
33
33
|
Requires-Dist: psutil==7.2.2
|
|
34
34
|
Requires-Dist: pyyaml==6.0.3
|
|
35
35
|
Provides-Extra: dev
|
|
36
36
|
Requires-Dist: pytest>=7.0; extra == "dev"
|
|
37
37
|
Requires-Dist: httpx>=0.27; extra == "dev"
|
|
38
|
-
Requires-Dist: ruff==0.16.
|
|
38
|
+
Requires-Dist: ruff==0.16.8; extra == "dev"
|
|
39
39
|
Requires-Dist: build>=1.2; extra == "dev"
|
|
40
40
|
Requires-Dist: twine>=5.0; extra == "dev"
|
|
41
41
|
Requires-Dist: playwright>=1.40; extra == "dev"
|
|
@@ -47,13 +47,13 @@ Requires-Dist: pillow>=10.0; extra == "packaging"
|
|
|
47
47
|
Dynamic: license-file
|
|
48
48
|
|
|
49
49
|
<p align="center">
|
|
50
|
-
<img src="localdeploy/web/logo.svg" width="96" height="96" alt="LocalDeploy logo" />
|
|
50
|
+
<img src="https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/localdeploy/web/logo.svg" width="96" height="96" alt="LocalDeploy logo" />
|
|
51
51
|
</p>
|
|
52
52
|
|
|
53
53
|
<h1 align="center">LocalDeploy</h1>
|
|
54
54
|
|
|
55
55
|
<p align="center">
|
|
56
|
-
|
|
56
|
+
<strong>Pick, deploy, and benchmark the best local AI model for your machine.</strong>
|
|
57
57
|
</p>
|
|
58
58
|
|
|
59
59
|
<p align="center">
|
|
@@ -64,14 +64,33 @@ Dynamic: license-file
|
|
|
64
64
|
<img src="https://img.shields.io/badge/platform-Windows%20%7C%20macOS%20%7C%20Linux%20%7C%20Docker-555.svg" alt="Platforms" />
|
|
65
65
|
</p>
|
|
66
66
|
|
|
67
|
-
I built LocalDeploy because I needed a dependable way to choose and run local models for other work I am involved in. It became useful enough that I decided to make it public. I hope it saves someone else some setup time and trial and error.
|
|
68
|
-
|
|
69
|
-
LocalDeploy sits on top of [Ollama](https://ollama.com) and can also work with llama.cpp and loopback OpenAI-compatible runtimes. It detects the machine, estimates whether a model will fit, manages local models, and records benchmark results. LocalDeploy itself does not provide cloud inference.
|
|
70
|
-
|
|
71
67
|
<p align="center">
|
|
72
|
-
<img src="docs/assets/demo.gif" alt="LocalDeploy hardware
|
|
68
|
+
<img src="https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/docs/assets/demo.gif" alt="LocalDeploy walkthrough showing hardware-aware model recommendations, deployment, local chat, benchmark comparisons, and live monitoring" width="820" />
|
|
73
69
|
</p>
|
|
74
70
|
|
|
71
|
+
LocalDeploy sits on top of [Ollama](https://ollama.com) and can also work with llama.cpp and loopback OpenAI-compatible runtimes. It detects your hardware, estimates whether a model will fit before you download it, manages local models, and records benchmark results. It serves an OpenAI-compatible API, so tools you already use can talk to the model you deployed. It has no telemetry and does not provide cloud inference.
|
|
72
|
+
|
|
73
|
+
## Quick start
|
|
74
|
+
|
|
75
|
+
With Python 3.10+ and Ollama already installed:
|
|
76
|
+
|
|
77
|
+
```bash
|
|
78
|
+
pip install localdeploy
|
|
79
|
+
localdeploy # starts the server and opens http://localhost:8000/ui
|
|
80
|
+
```
|
|
81
|
+
|
|
82
|
+
`uvx localdeploy` runs it without installing, if you use [uv](https://docs.astral.sh/uv/). Add `--no-browser` to keep the browser closed.
|
|
83
|
+
|
|
84
|
+
Without them, use the launcher from a clone. It creates the environment, opens the UI, and on macOS and Linux offers to install a missing Python or Ollama first:
|
|
85
|
+
|
|
86
|
+
```bash
|
|
87
|
+
git clone https://github.com/oney-erge/LocalDeploy.git
|
|
88
|
+
cd LocalDeploy
|
|
89
|
+
./run.sh # Linux. macOS: ./run.command Windows: .\run.bat
|
|
90
|
+
```
|
|
91
|
+
|
|
92
|
+
Docker, the Windows installer, and the details are under [Install](#install).
|
|
93
|
+
|
|
75
94
|
## What it does
|
|
76
95
|
|
|
77
96
|
- Detects NVIDIA, AMD, Intel, and Apple Silicon hardware, including compatible multi-GPU layouts.
|
|
@@ -86,8 +105,37 @@ LocalDeploy sits on top of [Ollama](https://ollama.com) and can also work with l
|
|
|
86
105
|
|
|
87
106
|
If you already know which Ollama model you want and only need a terminal chat, Ollama may be all you need. LocalDeploy is meant for the less certain part: choosing a model for a particular machine and comparing it with real runs.
|
|
88
107
|
|
|
108
|
+
## Why it exists
|
|
109
|
+
|
|
110
|
+
I built LocalDeploy because I needed a dependable way to choose and run local models for other work I am involved in. It became useful enough that I decided to make it public. I hope it saves someone else some setup time and trial and error.
|
|
111
|
+
|
|
89
112
|
## Install
|
|
90
113
|
|
|
114
|
+
The stable project entry points are `run.bat` on Windows, `run.command` on
|
|
115
|
+
macOS, and `run.sh` on Linux. Re-running them is safe. They use the existing
|
|
116
|
+
environment when it is current and repair missing dependencies when needed.
|
|
117
|
+
Each also accepts `doctor`, `repair`, `docker`, `logs`, and `stop`.
|
|
118
|
+
|
|
119
|
+
Setup checks disk space, prevents concurrent environment changes, retries
|
|
120
|
+
temporary network failures up to three times, and writes persistent failures
|
|
121
|
+
to `.setup/install.log`. The `logs` action follows the active native or
|
|
122
|
+
Docker logs rather than requiring Docker.
|
|
123
|
+
|
|
124
|
+
From a clone:
|
|
125
|
+
|
|
126
|
+
```powershell
|
|
127
|
+
.\run.bat # Windows
|
|
128
|
+
```
|
|
129
|
+
|
|
130
|
+
```bash
|
|
131
|
+
./run.command # macOS
|
|
132
|
+
./run.sh # Linux
|
|
133
|
+
```
|
|
134
|
+
|
|
135
|
+
Use `.\run.ps1` from PowerShell. Add an action after any launcher, such as
|
|
136
|
+
`.\run.ps1 doctor` or `./run.sh docker`. The default `run` action waits for
|
|
137
|
+
LocalDeploy to become ready before opening the UI.
|
|
138
|
+
|
|
91
139
|
Install from PyPI when Python and Ollama are already available:
|
|
92
140
|
|
|
93
141
|
```bash
|
|
@@ -107,27 +155,28 @@ Invoke-RestMethod https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/s
|
|
|
107
155
|
& $installer
|
|
108
156
|
```
|
|
109
157
|
|
|
110
|
-
From an existing clone:
|
|
158
|
+
From an existing clone, use the stable root launcher:
|
|
111
159
|
|
|
112
160
|
```powershell
|
|
113
161
|
git clone https://github.com/oney-erge/LocalDeploy.git
|
|
114
162
|
cd LocalDeploy
|
|
115
|
-
.\
|
|
163
|
+
.\run.bat
|
|
116
164
|
```
|
|
117
165
|
|
|
118
|
-
The
|
|
166
|
+
The launcher creates `.env`, `config.json`, and `.venv`, starts Ollama when it is installed, and opens `http://localhost:8000/ui`. Stop it with `.\run.ps1 stop`. The lower-level `scripts\start.ps1` remains available for development-specific flags.
|
|
119
167
|
|
|
120
168
|
### macOS and Linux
|
|
121
169
|
|
|
122
170
|
```bash
|
|
123
171
|
git clone https://github.com/oney-erge/LocalDeploy.git
|
|
124
172
|
cd LocalDeploy
|
|
125
|
-
./
|
|
173
|
+
./run.command # macOS
|
|
174
|
+
./run.sh # Linux
|
|
126
175
|
```
|
|
127
176
|
|
|
128
177
|
If Python 3 or Ollama is missing, the launcher offers to install it for you (Homebrew on macOS; apt, dnf, pacman, or zypper for Python and the [official installer](https://ollama.com/download) for Ollama on Linux) and asks before running anything. Answer no, or run non-interactively, and it prints the exact manual command instead.
|
|
129
178
|
|
|
130
|
-
On macOS you can
|
|
179
|
+
On macOS you can double-click `run.command` from a clone. A ZIP download can strip the executable bit, so after unzipping run `chmod +x run.command run.sh scripts/*.sh` once, or use `git clone` as above.
|
|
131
180
|
|
|
132
181
|
The launcher starts Ollama when it is installed but not already reachable, then keeps the LocalDeploy server in the foreground. Press Ctrl+C to stop LocalDeploy. Run `./scripts/stop.sh --ollama` if you also want to stop an Ollama process started by the launcher. Set `START_OLLAMA=false` in `.env` when another service manages Ollama.
|
|
133
182
|
|
|
@@ -140,7 +189,7 @@ docker run -d --name localdeploy \
|
|
|
140
189
|
-p 127.0.0.1:8000:8000 \
|
|
141
190
|
-v localdeploy-data:/data/localdeploy \
|
|
142
191
|
-v ollama-data:/root/.ollama \
|
|
143
|
-
ghcr.io/oney-erge/localdeploy:
|
|
192
|
+
ghcr.io/oney-erge/localdeploy:latest
|
|
144
193
|
```
|
|
145
194
|
|
|
146
195
|
To build from source instead:
|
|
@@ -162,11 +211,11 @@ A new install starts without profiles or models. Pulling the first model creates
|
|
|
162
211
|
The hardware and memory estimates are guidance. Drivers, runtime versions, context length, desktop GPU use, and quantization all affect the actual result. LocalDeploy records observed memory after deployment so later estimates can be adjusted for that machine.
|
|
163
212
|
|
|
164
213
|
<p align="center">
|
|
165
|
-
<img src="docs/screenshots/setup-deploy.png" alt="Setup and Deploy tab with hardware detection, live VRAM, and model fit information" width="800" />
|
|
214
|
+
<img src="https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/docs/screenshots/setup-deploy.png" alt="Setup and Deploy tab with hardware detection, live VRAM, and model fit information" width="800" />
|
|
166
215
|
</p>
|
|
167
216
|
|
|
168
217
|
<p align="center">
|
|
169
|
-
<img src="docs/screenshots/chat-playground.png" alt="Chat playground with a streamed response from a local model" width="800" />
|
|
218
|
+
<img src="https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/docs/screenshots/chat-playground.png" alt="Chat playground with a streamed response from a local model" width="800" />
|
|
170
219
|
</p>
|
|
171
220
|
|
|
172
221
|
More screenshots are in [docs/SCREENSHOTS.md](docs/SCREENSHOTS.md).
|
|
@@ -1,11 +1,11 @@
|
|
|
1
1
|
<p align="center">
|
|
2
|
-
<img src="localdeploy/web/logo.svg" width="96" height="96" alt="LocalDeploy logo" />
|
|
2
|
+
<img src="https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/localdeploy/web/logo.svg" width="96" height="96" alt="LocalDeploy logo" />
|
|
3
3
|
</p>
|
|
4
4
|
|
|
5
5
|
<h1 align="center">LocalDeploy</h1>
|
|
6
6
|
|
|
7
7
|
<p align="center">
|
|
8
|
-
|
|
8
|
+
<strong>Pick, deploy, and benchmark the best local AI model for your machine.</strong>
|
|
9
9
|
</p>
|
|
10
10
|
|
|
11
11
|
<p align="center">
|
|
@@ -16,14 +16,33 @@
|
|
|
16
16
|
<img src="https://img.shields.io/badge/platform-Windows%20%7C%20macOS%20%7C%20Linux%20%7C%20Docker-555.svg" alt="Platforms" />
|
|
17
17
|
</p>
|
|
18
18
|
|
|
19
|
-
I built LocalDeploy because I needed a dependable way to choose and run local models for other work I am involved in. It became useful enough that I decided to make it public. I hope it saves someone else some setup time and trial and error.
|
|
20
|
-
|
|
21
|
-
LocalDeploy sits on top of [Ollama](https://ollama.com) and can also work with llama.cpp and loopback OpenAI-compatible runtimes. It detects the machine, estimates whether a model will fit, manages local models, and records benchmark results. LocalDeploy itself does not provide cloud inference.
|
|
22
|
-
|
|
23
19
|
<p align="center">
|
|
24
|
-
<img src="docs/assets/demo.gif" alt="LocalDeploy hardware
|
|
20
|
+
<img src="https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/docs/assets/demo.gif" alt="LocalDeploy walkthrough showing hardware-aware model recommendations, deployment, local chat, benchmark comparisons, and live monitoring" width="820" />
|
|
25
21
|
</p>
|
|
26
22
|
|
|
23
|
+
LocalDeploy sits on top of [Ollama](https://ollama.com) and can also work with llama.cpp and loopback OpenAI-compatible runtimes. It detects your hardware, estimates whether a model will fit before you download it, manages local models, and records benchmark results. It serves an OpenAI-compatible API, so tools you already use can talk to the model you deployed. It has no telemetry and does not provide cloud inference.
|
|
24
|
+
|
|
25
|
+
## Quick start
|
|
26
|
+
|
|
27
|
+
With Python 3.10+ and Ollama already installed:
|
|
28
|
+
|
|
29
|
+
```bash
|
|
30
|
+
pip install localdeploy
|
|
31
|
+
localdeploy # starts the server and opens http://localhost:8000/ui
|
|
32
|
+
```
|
|
33
|
+
|
|
34
|
+
`uvx localdeploy` runs it without installing, if you use [uv](https://docs.astral.sh/uv/). Add `--no-browser` to keep the browser closed.
|
|
35
|
+
|
|
36
|
+
Without them, use the launcher from a clone. It creates the environment, opens the UI, and on macOS and Linux offers to install a missing Python or Ollama first:
|
|
37
|
+
|
|
38
|
+
```bash
|
|
39
|
+
git clone https://github.com/oney-erge/LocalDeploy.git
|
|
40
|
+
cd LocalDeploy
|
|
41
|
+
./run.sh # Linux. macOS: ./run.command Windows: .\run.bat
|
|
42
|
+
```
|
|
43
|
+
|
|
44
|
+
Docker, the Windows installer, and the details are under [Install](#install).
|
|
45
|
+
|
|
27
46
|
## What it does
|
|
28
47
|
|
|
29
48
|
- Detects NVIDIA, AMD, Intel, and Apple Silicon hardware, including compatible multi-GPU layouts.
|
|
@@ -38,8 +57,37 @@ LocalDeploy sits on top of [Ollama](https://ollama.com) and can also work with l
|
|
|
38
57
|
|
|
39
58
|
If you already know which Ollama model you want and only need a terminal chat, Ollama may be all you need. LocalDeploy is meant for the less certain part: choosing a model for a particular machine and comparing it with real runs.
|
|
40
59
|
|
|
60
|
+
## Why it exists
|
|
61
|
+
|
|
62
|
+
I built LocalDeploy because I needed a dependable way to choose and run local models for other work I am involved in. It became useful enough that I decided to make it public. I hope it saves someone else some setup time and trial and error.
|
|
63
|
+
|
|
41
64
|
## Install
|
|
42
65
|
|
|
66
|
+
The stable project entry points are `run.bat` on Windows, `run.command` on
|
|
67
|
+
macOS, and `run.sh` on Linux. Re-running them is safe. They use the existing
|
|
68
|
+
environment when it is current and repair missing dependencies when needed.
|
|
69
|
+
Each also accepts `doctor`, `repair`, `docker`, `logs`, and `stop`.
|
|
70
|
+
|
|
71
|
+
Setup checks disk space, prevents concurrent environment changes, retries
|
|
72
|
+
temporary network failures up to three times, and writes persistent failures
|
|
73
|
+
to `.setup/install.log`. The `logs` action follows the active native or
|
|
74
|
+
Docker logs rather than requiring Docker.
|
|
75
|
+
|
|
76
|
+
From a clone:
|
|
77
|
+
|
|
78
|
+
```powershell
|
|
79
|
+
.\run.bat # Windows
|
|
80
|
+
```
|
|
81
|
+
|
|
82
|
+
```bash
|
|
83
|
+
./run.command # macOS
|
|
84
|
+
./run.sh # Linux
|
|
85
|
+
```
|
|
86
|
+
|
|
87
|
+
Use `.\run.ps1` from PowerShell. Add an action after any launcher, such as
|
|
88
|
+
`.\run.ps1 doctor` or `./run.sh docker`. The default `run` action waits for
|
|
89
|
+
LocalDeploy to become ready before opening the UI.
|
|
90
|
+
|
|
43
91
|
Install from PyPI when Python and Ollama are already available:
|
|
44
92
|
|
|
45
93
|
```bash
|
|
@@ -59,27 +107,28 @@ Invoke-RestMethod https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/s
|
|
|
59
107
|
& $installer
|
|
60
108
|
```
|
|
61
109
|
|
|
62
|
-
From an existing clone:
|
|
110
|
+
From an existing clone, use the stable root launcher:
|
|
63
111
|
|
|
64
112
|
```powershell
|
|
65
113
|
git clone https://github.com/oney-erge/LocalDeploy.git
|
|
66
114
|
cd LocalDeploy
|
|
67
|
-
.\
|
|
115
|
+
.\run.bat
|
|
68
116
|
```
|
|
69
117
|
|
|
70
|
-
The
|
|
118
|
+
The launcher creates `.env`, `config.json`, and `.venv`, starts Ollama when it is installed, and opens `http://localhost:8000/ui`. Stop it with `.\run.ps1 stop`. The lower-level `scripts\start.ps1` remains available for development-specific flags.
|
|
71
119
|
|
|
72
120
|
### macOS and Linux
|
|
73
121
|
|
|
74
122
|
```bash
|
|
75
123
|
git clone https://github.com/oney-erge/LocalDeploy.git
|
|
76
124
|
cd LocalDeploy
|
|
77
|
-
./
|
|
125
|
+
./run.command # macOS
|
|
126
|
+
./run.sh # Linux
|
|
78
127
|
```
|
|
79
128
|
|
|
80
129
|
If Python 3 or Ollama is missing, the launcher offers to install it for you (Homebrew on macOS; apt, dnf, pacman, or zypper for Python and the [official installer](https://ollama.com/download) for Ollama on Linux) and asks before running anything. Answer no, or run non-interactively, and it prints the exact manual command instead.
|
|
81
130
|
|
|
82
|
-
On macOS you can
|
|
131
|
+
On macOS you can double-click `run.command` from a clone. A ZIP download can strip the executable bit, so after unzipping run `chmod +x run.command run.sh scripts/*.sh` once, or use `git clone` as above.
|
|
83
132
|
|
|
84
133
|
The launcher starts Ollama when it is installed but not already reachable, then keeps the LocalDeploy server in the foreground. Press Ctrl+C to stop LocalDeploy. Run `./scripts/stop.sh --ollama` if you also want to stop an Ollama process started by the launcher. Set `START_OLLAMA=false` in `.env` when another service manages Ollama.
|
|
85
134
|
|
|
@@ -92,7 +141,7 @@ docker run -d --name localdeploy \
|
|
|
92
141
|
-p 127.0.0.1:8000:8000 \
|
|
93
142
|
-v localdeploy-data:/data/localdeploy \
|
|
94
143
|
-v ollama-data:/root/.ollama \
|
|
95
|
-
ghcr.io/oney-erge/localdeploy:
|
|
144
|
+
ghcr.io/oney-erge/localdeploy:latest
|
|
96
145
|
```
|
|
97
146
|
|
|
98
147
|
To build from source instead:
|
|
@@ -114,11 +163,11 @@ A new install starts without profiles or models. Pulling the first model creates
|
|
|
114
163
|
The hardware and memory estimates are guidance. Drivers, runtime versions, context length, desktop GPU use, and quantization all affect the actual result. LocalDeploy records observed memory after deployment so later estimates can be adjusted for that machine.
|
|
115
164
|
|
|
116
165
|
<p align="center">
|
|
117
|
-
<img src="docs/screenshots/setup-deploy.png" alt="Setup and Deploy tab with hardware detection, live VRAM, and model fit information" width="800" />
|
|
166
|
+
<img src="https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/docs/screenshots/setup-deploy.png" alt="Setup and Deploy tab with hardware detection, live VRAM, and model fit information" width="800" />
|
|
118
167
|
</p>
|
|
119
168
|
|
|
120
169
|
<p align="center">
|
|
121
|
-
<img src="docs/screenshots/chat-playground.png" alt="Chat playground with a streamed response from a local model" width="800" />
|
|
170
|
+
<img src="https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/docs/screenshots/chat-playground.png" alt="Chat playground with a streamed response from a local model" width="800" />
|
|
122
171
|
</p>
|
|
123
172
|
|
|
124
173
|
More screenshots are in [docs/SCREENSHOTS.md](docs/SCREENSHOTS.md).
|
|
@@ -646,7 +646,7 @@
|
|
|
646
646
|
<div class="row gap">
|
|
647
647
|
<button class="btn danger" id="btn-run-cancel" hidden>Cancel</button>
|
|
648
648
|
<button class="btn" id="btn-export" disabled title="Select a completed run">Export run</button>
|
|
649
|
-
<button class="btn" id="btn-contribute" disabled title="Select a completed run">
|
|
649
|
+
<button class="btn" id="btn-contribute" disabled title="Select a completed run">Save anonymized benchmark</button>
|
|
650
650
|
</div>
|
|
651
651
|
</div>
|
|
652
652
|
<div class="bench-summary-grid">
|
|
@@ -18,7 +18,7 @@ export function renderWinners(runs) {
|
|
|
18
18
|
const bestTps = [...runs].filter((run) => run.summary.avg_tokens_per_second != null).sort((a, b) => b.summary.avg_tokens_per_second - a.summary.avg_tokens_per_second)[0];
|
|
19
19
|
const categories = new Set(runs.flatMap((run) => (run.category_summary || categorySummary(run.tests)).map((row) => row.category)));
|
|
20
20
|
return [
|
|
21
|
-
`<div class="metric-tile"><span>Top run</span><strong>${esc(runLabel(top))}</strong><small>${esc(top.summary.passed)}/${esc(top.summary.tests)}
|
|
21
|
+
`<div class="metric-tile"><span>Top run</span><strong>${esc(runLabel(top))}</strong><small>${esc(top.summary.passed)}/${esc(top.summary.tests)} completed</small></div>`,
|
|
22
22
|
`<div class="metric-tile"><span>Accuracy</span><strong>${esc(top.summary.avg_accuracy)}</strong><small>${esc(top.questionSetName || "benchmark")}</small></div>`,
|
|
23
23
|
fastest ? `<div class="metric-tile"><span>Fastest</span><strong>${esc(fastest.summary.avg_latency_s)}s</strong><small>${esc(runLabel(fastest))}</small></div>` : "",
|
|
24
24
|
bestTps ? `<div class="metric-tile"><span>Best tok/s</span><strong>${esc(bestTps.summary.avg_tokens_per_second)}</strong><small>${esc(runLabel(bestTps))}</small></div>` : "",
|
|
@@ -28,7 +28,7 @@ export function renderWinners(runs) {
|
|
|
28
28
|
|
|
29
29
|
export function renderLeaderboard(runs) {
|
|
30
30
|
if (!runs.length) return "Benchmark results appear here after the first streamed test result.";
|
|
31
|
-
const rows = [...runs].sort((a, b) => (b.summary.passed || 0) - (a.summary.passed || 0) || (b.summary.avg_accuracy || 0) - (a.summary.avg_accuracy || 0) || (a.summary.avg_latency_s || 999999) - (b.summary.avg_latency_s || 999999)).map((run, index) => `<div class="leaderboard-row"><span class="rank">${index + 1}</span><div class="leaderboard-name"><b>${esc(runLabel(run))}</b><span>${esc(run.questionSetName || "")}</span></div><div class="leaderboard-metrics"><span><b>${esc(run.summary.passed)}/${esc(run.summary.tests)}</b>
|
|
31
|
+
const rows = [...runs].sort((a, b) => (b.summary.passed || 0) - (a.summary.passed || 0) || (b.summary.avg_accuracy || 0) - (a.summary.avg_accuracy || 0) || (a.summary.avg_latency_s || 999999) - (b.summary.avg_latency_s || 999999)).map((run, index) => `<div class="leaderboard-row"><span class="rank">${index + 1}</span><div class="leaderboard-name"><b>${esc(runLabel(run))}</b><span>${esc(run.questionSetName || "")}</span></div><div class="leaderboard-metrics"><span><b>${esc(run.summary.passed)}/${esc(run.summary.tests)}</b> completed</span><span><b>${esc(run.summary.avg_accuracy)}</b> acc</span><span><b>${esc(run.summary.avg_latency_s)}s</b> latency</span><span><b>${esc(run.summary.avg_tokens_per_second ?? "-")}</b> tok/s</span></div></div>`).join("");
|
|
32
32
|
return `<div class="leaderboard-list">${rows}</div>`;
|
|
33
33
|
}
|
|
34
34
|
|
|
@@ -1213,7 +1213,7 @@ async function renderComparison(runs, baseline) {
|
|
|
1213
1213
|
.map((name) => `<button class="btn compact response-compare-btn" data-test="${esc(name)}">${esc(name)}</button>`)
|
|
1214
1214
|
.join("");
|
|
1215
1215
|
$("#compare-body").innerHTML = `<div class="table-wrap"><table class="results">
|
|
1216
|
-
<thead><tr><th>Run</th><th class="num">
|
|
1216
|
+
<thead><tr><th>Run</th><th class="num">Completed</th><th class="num">Accuracy Δ</th><th class="num">Latency Δ</th><th class="num">tok/s Δ</th></tr></thead>
|
|
1217
1217
|
<tbody>${rows}</tbody></table></div>
|
|
1218
1218
|
<h3 class="sub">Response detail</h3><div class="response-test-list">${testButtons}</div>
|
|
1219
1219
|
<div id="regression-diffs"></div>`;
|
|
@@ -1329,8 +1329,8 @@ async function contributeRun() {
|
|
|
1329
1329
|
const out = await postJSON("/system/community/preview", { card: runToCardPayload(run) });
|
|
1330
1330
|
if (!out.success) throw new Error(out.error || "Preview failed.");
|
|
1331
1331
|
const overlay = simpleModal(
|
|
1332
|
-
"
|
|
1333
|
-
`
|
|
1332
|
+
"Save an anonymized benchmark",
|
|
1333
|
+
`This anonymized snapshot is saved to a file on this computer. Nothing is sent anywhere. ${esc(out.note)}`,
|
|
1334
1334
|
`<pre class="log manifest-yaml">${esc(JSON.stringify(out.would_share, null, 2))}</pre>
|
|
1335
1335
|
<details style="margin-top:0.5rem"><summary class="muted small">Never included</summary>
|
|
1336
1336
|
<div class="muted small">${out.excluded_fields.map(esc).join(", ")}</div>
|
|
@@ -1375,7 +1375,7 @@ function renderBakeoffResult(evt) {
|
|
|
1375
1375
|
<h3 class="sub">${esc(evt.winner)}</h3>
|
|
1376
1376
|
</div>
|
|
1377
1377
|
<div class="table-wrap"><table class="results">
|
|
1378
|
-
<thead><tr><th>Model</th><th class="num">
|
|
1378
|
+
<thead><tr><th>Model</th><th class="num">Completed</th><th class="num">Accuracy</th><th class="num">Avg latency</th><th class="num">Headroom</th></tr></thead>
|
|
1379
1379
|
<tbody>${rows}</tbody>
|
|
1380
1380
|
</table></div>
|
|
1381
1381
|
<div class="row gap wrap" style="margin-top:0.5rem">
|
|
@@ -1454,7 +1454,7 @@ table.results td.num, table.results th.num { text-align: right; font-variant-num
|
|
|
1454
1454
|
font-weight: 700;
|
|
1455
1455
|
}
|
|
1456
1456
|
.heat-category {
|
|
1457
|
-
color:
|
|
1457
|
+
color: var(--text);
|
|
1458
1458
|
font-weight: 650;
|
|
1459
1459
|
}
|
|
1460
1460
|
.heat-cell {
|
|
@@ -1462,9 +1462,9 @@ table.results td.num, table.results th.num { text-align: right; font-variant-num
|
|
|
1462
1462
|
font-variant-numeric: tabular-nums;
|
|
1463
1463
|
font-weight: 700;
|
|
1464
1464
|
}
|
|
1465
|
-
.heat-good { background: rgba(34, 133, 83, 0.48); color:
|
|
1466
|
-
.heat-warn { background: rgba(151, 122, 35, 0.42); color:
|
|
1467
|
-
.heat-bad { background: rgba(143, 55, 55, 0.38); color:
|
|
1465
|
+
.heat-good { background: rgba(34, 133, 83, 0.48); color: var(--text); }
|
|
1466
|
+
.heat-warn { background: rgba(151, 122, 35, 0.42); color: var(--text); }
|
|
1467
|
+
.heat-bad { background: rgba(143, 55, 55, 0.38); color: var(--text); }
|
|
1468
1468
|
.matrix-wrap {
|
|
1469
1469
|
max-height: 320px;
|
|
1470
1470
|
overflow: auto;
|
|
@@ -1494,8 +1494,8 @@ table.results td.num, table.results th.num { text-align: right; font-variant-num
|
|
|
1494
1494
|
font-weight: 750;
|
|
1495
1495
|
font-variant-numeric: tabular-nums;
|
|
1496
1496
|
}
|
|
1497
|
-
.matrix-pass { color:
|
|
1498
|
-
.matrix-fail { color:
|
|
1497
|
+
.matrix-pass { color: var(--text); background: rgba(34, 133, 83, 0.32); }
|
|
1498
|
+
.matrix-fail { color: var(--text); background: rgba(143, 55, 55, 0.3); }
|
|
1499
1499
|
.matrix-note { margin-top: 0.45rem; }
|
|
1500
1500
|
.run-library {
|
|
1501
1501
|
display: grid;
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: localdeploy
|
|
3
|
-
Version: 0.6.
|
|
4
|
-
Summary:
|
|
3
|
+
Version: 0.6.2
|
|
4
|
+
Summary: Find, deploy, manage, and benchmark local AI models for your hardware across Ollama, llama.cpp, and compatible runtimes.
|
|
5
5
|
Author: oney-erge
|
|
6
6
|
License-Expression: MIT
|
|
7
7
|
Project-URL: Homepage, https://github.com/oney-erge/LocalDeploy
|
|
@@ -23,19 +23,19 @@ Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
|
23
23
|
Requires-Python: >=3.10
|
|
24
24
|
Description-Content-Type: text/markdown
|
|
25
25
|
License-File: LICENSE
|
|
26
|
-
Requires-Dist: fastapi==0.
|
|
27
|
-
Requires-Dist: uvicorn==0.
|
|
28
|
-
Requires-Dist: requests==2.
|
|
29
|
-
Requires-Dist: python-dotenv==1.2.
|
|
30
|
-
Requires-Dist: pydantic==2.13.
|
|
31
|
-
Requires-Dist: python-multipart==0.0.
|
|
32
|
-
Requires-Dist: huggingface_hub==1.
|
|
26
|
+
Requires-Dist: fastapi==0.141.1
|
|
27
|
+
Requires-Dist: uvicorn==0.53.0
|
|
28
|
+
Requires-Dist: requests==2.34.2
|
|
29
|
+
Requires-Dist: python-dotenv==1.2.3
|
|
30
|
+
Requires-Dist: pydantic==2.13.5
|
|
31
|
+
Requires-Dist: python-multipart==0.0.32
|
|
32
|
+
Requires-Dist: huggingface_hub==1.31.0
|
|
33
33
|
Requires-Dist: psutil==7.2.2
|
|
34
34
|
Requires-Dist: pyyaml==6.0.3
|
|
35
35
|
Provides-Extra: dev
|
|
36
36
|
Requires-Dist: pytest>=7.0; extra == "dev"
|
|
37
37
|
Requires-Dist: httpx>=0.27; extra == "dev"
|
|
38
|
-
Requires-Dist: ruff==0.16.
|
|
38
|
+
Requires-Dist: ruff==0.16.8; extra == "dev"
|
|
39
39
|
Requires-Dist: build>=1.2; extra == "dev"
|
|
40
40
|
Requires-Dist: twine>=5.0; extra == "dev"
|
|
41
41
|
Requires-Dist: playwright>=1.40; extra == "dev"
|
|
@@ -47,13 +47,13 @@ Requires-Dist: pillow>=10.0; extra == "packaging"
|
|
|
47
47
|
Dynamic: license-file
|
|
48
48
|
|
|
49
49
|
<p align="center">
|
|
50
|
-
<img src="localdeploy/web/logo.svg" width="96" height="96" alt="LocalDeploy logo" />
|
|
50
|
+
<img src="https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/localdeploy/web/logo.svg" width="96" height="96" alt="LocalDeploy logo" />
|
|
51
51
|
</p>
|
|
52
52
|
|
|
53
53
|
<h1 align="center">LocalDeploy</h1>
|
|
54
54
|
|
|
55
55
|
<p align="center">
|
|
56
|
-
|
|
56
|
+
<strong>Pick, deploy, and benchmark the best local AI model for your machine.</strong>
|
|
57
57
|
</p>
|
|
58
58
|
|
|
59
59
|
<p align="center">
|
|
@@ -64,14 +64,33 @@ Dynamic: license-file
|
|
|
64
64
|
<img src="https://img.shields.io/badge/platform-Windows%20%7C%20macOS%20%7C%20Linux%20%7C%20Docker-555.svg" alt="Platforms" />
|
|
65
65
|
</p>
|
|
66
66
|
|
|
67
|
-
I built LocalDeploy because I needed a dependable way to choose and run local models for other work I am involved in. It became useful enough that I decided to make it public. I hope it saves someone else some setup time and trial and error.
|
|
68
|
-
|
|
69
|
-
LocalDeploy sits on top of [Ollama](https://ollama.com) and can also work with llama.cpp and loopback OpenAI-compatible runtimes. It detects the machine, estimates whether a model will fit, manages local models, and records benchmark results. LocalDeploy itself does not provide cloud inference.
|
|
70
|
-
|
|
71
67
|
<p align="center">
|
|
72
|
-
<img src="docs/assets/demo.gif" alt="LocalDeploy hardware
|
|
68
|
+
<img src="https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/docs/assets/demo.gif" alt="LocalDeploy walkthrough showing hardware-aware model recommendations, deployment, local chat, benchmark comparisons, and live monitoring" width="820" />
|
|
73
69
|
</p>
|
|
74
70
|
|
|
71
|
+
LocalDeploy sits on top of [Ollama](https://ollama.com) and can also work with llama.cpp and loopback OpenAI-compatible runtimes. It detects your hardware, estimates whether a model will fit before you download it, manages local models, and records benchmark results. It serves an OpenAI-compatible API, so tools you already use can talk to the model you deployed. It has no telemetry and does not provide cloud inference.
|
|
72
|
+
|
|
73
|
+
## Quick start
|
|
74
|
+
|
|
75
|
+
With Python 3.10+ and Ollama already installed:
|
|
76
|
+
|
|
77
|
+
```bash
|
|
78
|
+
pip install localdeploy
|
|
79
|
+
localdeploy # starts the server and opens http://localhost:8000/ui
|
|
80
|
+
```
|
|
81
|
+
|
|
82
|
+
`uvx localdeploy` runs it without installing, if you use [uv](https://docs.astral.sh/uv/). Add `--no-browser` to keep the browser closed.
|
|
83
|
+
|
|
84
|
+
Without them, use the launcher from a clone. It creates the environment, opens the UI, and on macOS and Linux offers to install a missing Python or Ollama first:
|
|
85
|
+
|
|
86
|
+
```bash
|
|
87
|
+
git clone https://github.com/oney-erge/LocalDeploy.git
|
|
88
|
+
cd LocalDeploy
|
|
89
|
+
./run.sh # Linux. macOS: ./run.command Windows: .\run.bat
|
|
90
|
+
```
|
|
91
|
+
|
|
92
|
+
Docker, the Windows installer, and the details are under [Install](#install).
|
|
93
|
+
|
|
75
94
|
## What it does
|
|
76
95
|
|
|
77
96
|
- Detects NVIDIA, AMD, Intel, and Apple Silicon hardware, including compatible multi-GPU layouts.
|
|
@@ -86,8 +105,37 @@ LocalDeploy sits on top of [Ollama](https://ollama.com) and can also work with l
|
|
|
86
105
|
|
|
87
106
|
If you already know which Ollama model you want and only need a terminal chat, Ollama may be all you need. LocalDeploy is meant for the less certain part: choosing a model for a particular machine and comparing it with real runs.
|
|
88
107
|
|
|
108
|
+
## Why it exists
|
|
109
|
+
|
|
110
|
+
I built LocalDeploy because I needed a dependable way to choose and run local models for other work I am involved in. It became useful enough that I decided to make it public. I hope it saves someone else some setup time and trial and error.
|
|
111
|
+
|
|
89
112
|
## Install
|
|
90
113
|
|
|
114
|
+
The stable project entry points are `run.bat` on Windows, `run.command` on
|
|
115
|
+
macOS, and `run.sh` on Linux. Re-running them is safe. They use the existing
|
|
116
|
+
environment when it is current and repair missing dependencies when needed.
|
|
117
|
+
Each also accepts `doctor`, `repair`, `docker`, `logs`, and `stop`.
|
|
118
|
+
|
|
119
|
+
Setup checks disk space, prevents concurrent environment changes, retries
|
|
120
|
+
temporary network failures up to three times, and writes persistent failures
|
|
121
|
+
to `.setup/install.log`. The `logs` action follows the active native or
|
|
122
|
+
Docker logs rather than requiring Docker.
|
|
123
|
+
|
|
124
|
+
From a clone:
|
|
125
|
+
|
|
126
|
+
```powershell
|
|
127
|
+
.\run.bat # Windows
|
|
128
|
+
```
|
|
129
|
+
|
|
130
|
+
```bash
|
|
131
|
+
./run.command # macOS
|
|
132
|
+
./run.sh # Linux
|
|
133
|
+
```
|
|
134
|
+
|
|
135
|
+
Use `.\run.ps1` from PowerShell. Add an action after any launcher, such as
|
|
136
|
+
`.\run.ps1 doctor` or `./run.sh docker`. The default `run` action waits for
|
|
137
|
+
LocalDeploy to become ready before opening the UI.
|
|
138
|
+
|
|
91
139
|
Install from PyPI when Python and Ollama are already available:
|
|
92
140
|
|
|
93
141
|
```bash
|
|
@@ -107,27 +155,28 @@ Invoke-RestMethod https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/s
|
|
|
107
155
|
& $installer
|
|
108
156
|
```
|
|
109
157
|
|
|
110
|
-
From an existing clone:
|
|
158
|
+
From an existing clone, use the stable root launcher:
|
|
111
159
|
|
|
112
160
|
```powershell
|
|
113
161
|
git clone https://github.com/oney-erge/LocalDeploy.git
|
|
114
162
|
cd LocalDeploy
|
|
115
|
-
.\
|
|
163
|
+
.\run.bat
|
|
116
164
|
```
|
|
117
165
|
|
|
118
|
-
The
|
|
166
|
+
The launcher creates `.env`, `config.json`, and `.venv`, starts Ollama when it is installed, and opens `http://localhost:8000/ui`. Stop it with `.\run.ps1 stop`. The lower-level `scripts\start.ps1` remains available for development-specific flags.
|
|
119
167
|
|
|
120
168
|
### macOS and Linux
|
|
121
169
|
|
|
122
170
|
```bash
|
|
123
171
|
git clone https://github.com/oney-erge/LocalDeploy.git
|
|
124
172
|
cd LocalDeploy
|
|
125
|
-
./
|
|
173
|
+
./run.command # macOS
|
|
174
|
+
./run.sh # Linux
|
|
126
175
|
```
|
|
127
176
|
|
|
128
177
|
If Python 3 or Ollama is missing, the launcher offers to install it for you (Homebrew on macOS; apt, dnf, pacman, or zypper for Python and the [official installer](https://ollama.com/download) for Ollama on Linux) and asks before running anything. Answer no, or run non-interactively, and it prints the exact manual command instead.
|
|
129
178
|
|
|
130
|
-
On macOS you can
|
|
179
|
+
On macOS you can double-click `run.command` from a clone. A ZIP download can strip the executable bit, so after unzipping run `chmod +x run.command run.sh scripts/*.sh` once, or use `git clone` as above.
|
|
131
180
|
|
|
132
181
|
The launcher starts Ollama when it is installed but not already reachable, then keeps the LocalDeploy server in the foreground. Press Ctrl+C to stop LocalDeploy. Run `./scripts/stop.sh --ollama` if you also want to stop an Ollama process started by the launcher. Set `START_OLLAMA=false` in `.env` when another service manages Ollama.
|
|
133
182
|
|
|
@@ -140,7 +189,7 @@ docker run -d --name localdeploy \
|
|
|
140
189
|
-p 127.0.0.1:8000:8000 \
|
|
141
190
|
-v localdeploy-data:/data/localdeploy \
|
|
142
191
|
-v ollama-data:/root/.ollama \
|
|
143
|
-
ghcr.io/oney-erge/localdeploy:
|
|
192
|
+
ghcr.io/oney-erge/localdeploy:latest
|
|
144
193
|
```
|
|
145
194
|
|
|
146
195
|
To build from source instead:
|
|
@@ -162,11 +211,11 @@ A new install starts without profiles or models. Pulling the first model creates
|
|
|
162
211
|
The hardware and memory estimates are guidance. Drivers, runtime versions, context length, desktop GPU use, and quantization all affect the actual result. LocalDeploy records observed memory after deployment so later estimates can be adjusted for that machine.
|
|
163
212
|
|
|
164
213
|
<p align="center">
|
|
165
|
-
<img src="docs/screenshots/setup-deploy.png" alt="Setup and Deploy tab with hardware detection, live VRAM, and model fit information" width="800" />
|
|
214
|
+
<img src="https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/docs/screenshots/setup-deploy.png" alt="Setup and Deploy tab with hardware detection, live VRAM, and model fit information" width="800" />
|
|
166
215
|
</p>
|
|
167
216
|
|
|
168
217
|
<p align="center">
|
|
169
|
-
<img src="docs/screenshots/chat-playground.png" alt="Chat playground with a streamed response from a local model" width="800" />
|
|
218
|
+
<img src="https://raw.githubusercontent.com/oney-erge/LocalDeploy/main/docs/screenshots/chat-playground.png" alt="Chat playground with a streamed response from a local model" width="800" />
|
|
170
219
|
</p>
|
|
171
220
|
|
|
172
221
|
More screenshots are in [docs/SCREENSHOTS.md](docs/SCREENSHOTS.md).
|
|
@@ -1,17 +1,17 @@
|
|
|
1
|
-
fastapi==0.
|
|
2
|
-
uvicorn==0.
|
|
3
|
-
requests==2.
|
|
4
|
-
python-dotenv==1.2.
|
|
5
|
-
pydantic==2.13.
|
|
6
|
-
python-multipart==0.0.
|
|
7
|
-
huggingface_hub==1.
|
|
1
|
+
fastapi==0.141.1
|
|
2
|
+
uvicorn==0.53.0
|
|
3
|
+
requests==2.34.2
|
|
4
|
+
python-dotenv==1.2.3
|
|
5
|
+
pydantic==2.13.5
|
|
6
|
+
python-multipart==0.0.32
|
|
7
|
+
huggingface_hub==1.31.0
|
|
8
8
|
psutil==7.2.2
|
|
9
9
|
pyyaml==6.0.3
|
|
10
10
|
|
|
11
11
|
[dev]
|
|
12
12
|
pytest>=7.0
|
|
13
13
|
httpx>=0.27
|
|
14
|
-
ruff==0.16.
|
|
14
|
+
ruff==0.16.8
|
|
15
15
|
build>=1.2
|
|
16
16
|
twine>=5.0
|
|
17
17
|
playwright>=1.40
|
|
@@ -5,7 +5,7 @@ build-backend = "setuptools.build_meta"
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "localdeploy"
|
|
7
7
|
dynamic = ["version", "dependencies"]
|
|
8
|
-
description = "
|
|
8
|
+
description = "Find, deploy, manage, and benchmark local AI models for your hardware across Ollama, llama.cpp, and compatible runtimes."
|
|
9
9
|
readme = "README.md"
|
|
10
10
|
license = "MIT"
|
|
11
11
|
license-files = ["LICENSE"]
|
|
@@ -29,7 +29,7 @@ classifiers = [
|
|
|
29
29
|
dev = [
|
|
30
30
|
"pytest>=7.0",
|
|
31
31
|
"httpx>=0.27",
|
|
32
|
-
"ruff==0.16.
|
|
32
|
+
"ruff==0.16.8",
|
|
33
33
|
"build>=1.2",
|
|
34
34
|
"twine>=5.0",
|
|
35
35
|
"playwright>=1.40",
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|