amagi-model-sunset 0.1.0__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.
@@ -0,0 +1,21 @@
1
+ MIT License
2
+
3
+ Copyright (c) 2026 Amagisolutionsgroup, Inc. (AmagiLabs)
4
+
5
+ Permission is hereby granted, free of charge, to any person obtaining a copy
6
+ of this software and associated documentation files (the "Software"), to deal
7
+ in the Software without restriction, including without limitation the rights
8
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9
+ copies of the Software, and to permit persons to whom the Software is
10
+ furnished to do so, subject to the following conditions:
11
+
12
+ The above copyright notice and this permission notice shall be included in all
13
+ copies or substantial portions of the Software.
14
+
15
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21
+ SOFTWARE.
@@ -0,0 +1,141 @@
1
+ Metadata-Version: 2.4
2
+ Name: amagi-model-sunset
3
+ Version: 0.1.0
4
+ Summary: Finds hard-coded LLM model ids in your repo and tells you which ones the provider is retiring, and when
5
+ Author-email: "AmagiLabs (Amagisolutionsgroup, Inc.)" <dev@amagisolutionsgroup.com>
6
+ License: MIT
7
+ Keywords: llm,openai,anthropic,gemini,deprecation,ci,cli
8
+ Classifier: Programming Language :: Python :: 3
9
+ Classifier: License :: OSI Approved :: MIT License
10
+ Classifier: Operating System :: OS Independent
11
+ Classifier: Environment :: Console
12
+ Requires-Python: >=3.9
13
+ Description-Content-Type: text/markdown
14
+ License-File: LICENSE
15
+ Provides-Extra: dev
16
+ Requires-Dist: pytest>=7.0; extra == "dev"
17
+ Dynamic: license-file
18
+
19
+ # amagi-model-sunset
20
+
21
+ Scans your repo for hard-coded LLM model ids, tells you which ones OpenAI, Anthropic or Google has retired or is about to retire, and suggests the replacement. Built for developers and small teams who would rather find out in CI than from a 404 in production.
22
+
23
+ ![amagi-model-sunset demo](assets/demo.gif)
24
+
25
+ ## Install
26
+
27
+ ```bash
28
+ pipx install amagi-model-sunset
29
+ ```
30
+
31
+ Python 3.9+, no dependencies. Nothing leaves your machine: it reads your files and a dated data file that ships with the package.
32
+
33
+ ## Use
34
+
35
+ Point it at a project:
36
+
37
+ ```bash
38
+ amagi-model-sunset .
39
+ ```
40
+
41
+ This run is against the demo app in the repo (`examples/demo-app`: a Python file, a YAML config and an `.env.example`). `--today` pins the date so the output below is reproducible:
42
+
43
+ ```bash
44
+ amagi-model-sunset examples/demo-app --today 2026-11-16
45
+ ```
46
+
47
+ ```text
48
+ config/llm.yaml:3 gemini-1.5-pro-002 -> RETIRED 2025-09-24 (418d ago)
49
+ suggest: gemini-3.6-flash
50
+ config/llm.yaml:4 gemini-1.5-pro-002 -> RETIRED 2025-09-24 (418d ago)
51
+ suggest: gemini-3.6-flash
52
+ config/llm.yaml:6 anthropic.claude-3-5-sonnet-20241022-v2:0 -> RETIRED 2025-10-28 (384d ago)
53
+ suggest: claude-sonnet-5-5
54
+ app.py:14 claude-3-opus-20240229 -> RETIRED 2026-01-05 (315d ago)
55
+ suggest: claude-opus-4-8
56
+ config/llm.yaml:7 us.anthropic.claude-3-opus-20240229-v1:0 -> RETIRED 2026-01-05 (315d ago)
57
+ suggest: claude-opus-4-8
58
+ .env.example:3 text-embedding-004 -> RETIRED 2026-01-14 (306d ago)
59
+ suggest: gemini-embedding-2
60
+ .env.example:2 gpt-4-turbo -> RETIRED 2026-10-23 (24d ago)
61
+ suggest: gpt-5.6-sol
62
+ app.py:6 gpt-4-0613 -> RETIRED 2026-10-23 (24d ago)
63
+ suggest: gpt-5.6-sol
64
+ config/llm.yaml:8 claude-sonnet-4-5-20250929 -> RETIRES 2026-11-30 (in 14d)
65
+ suggest: claude-sonnet-5-5
66
+ app.py:7 gpt-4o-mini -> RETIRES 2027-01-14 (in 59d)
67
+ suggest: gpt-6-luna
68
+
69
+ source: https://cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions (checked 2026-11-16)
70
+ source: https://platform.claude.com/docs/en/about-claude/model-deprecations (checked 2026-11-16)
71
+ source: https://ai.google.dev/gemini-api/docs/deprecations (checked 2026-11-16)
72
+ source: https://platform.openai.com/docs/deprecations (checked 2026-11-16)
73
+ scanned examples/demo-app: 8 model id(s) in 10 place(s): 6 retired, 2 retiring, 0 deprecated without a date, 0 ok
74
+ FAIL: 9 place(s) use a model that is retired or retires within 30 days
75
+ data checked 2026-11-16 against the providers' own pages; run with --format json for sources
76
+ ```
77
+
78
+ It finds plain ids (`gpt-4-0613`), env defaults, YAML values, Bedrock ids and ARNs (`us.anthropic.claude-3-opus-20240229-v1:0`), Vertex paths and `@date` ids, and fine-tune ids (`ft:gpt-4o-mini-...`).
79
+
80
+ ## Fail CI before the 404s start
81
+
82
+ ```bash
83
+ amagi-model-sunset . --fail-within 30
84
+ ```
85
+
86
+ Exit code 1 when a model is already retired or retires within 30 days, 0 otherwise. Use `--fail-within 0` to fail only on models that are already gone. A GitHub Actions step:
87
+
88
+ ```yaml
89
+ - run: pipx install amagi-model-sunset
90
+ - run: amagi-model-sunset . --fail-within 30 --exclude "docs/*"
91
+ ```
92
+
93
+ Dates change, so run it on a weekly schedule as well as on pull requests.
94
+
95
+ More options:
96
+
97
+ ```bash
98
+ amagi-model-sunset . --format json # machine-readable, with source URL and checked-on date
99
+ amagi-model-sunset . --format md # table for a PR comment
100
+ amagi-model-sunset . --all # also list models with no announced retirement
101
+ amagi-model-sunset . --exclude "*.md" # skip files (repeatable)
102
+ amagi-model-sunset . --list-models # every id the data file knows
103
+ amagi-model-sunset . --data my-models.json # use your own or a newer data file
104
+ ```
105
+
106
+ Add `# model-sunset: ignore` to a line to skip it (a migration note, a test that checks old ids).
107
+
108
+ ## Where the dates come from
109
+
110
+ `src/amagi_model_sunset/data/models.json` holds about 245 ids. Each entry carries the provider page it was copied from and the date it was checked (2026-11-16, or 2026-11-20 for rows added that day):
111
+
112
+ - OpenAI: https://platform.openai.com/docs/deprecations
113
+ - Anthropic: https://platform.claude.com/docs/en/about-claude/model-deprecations
114
+ - Google Gemini API: https://ai.google.dev/gemini-api/docs/deprecations
115
+ - Google Vertex AI (Gemini 1.x): https://cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions
116
+
117
+ Limits to know about:
118
+
119
+ - Entries marked `verified: false` were not confirmed on the provider page (older aliases that dropped off it). The tool labels them `[unverified date]`.
120
+ - "Suggest" is the provider's own recommendation, not a drop-in guarantee. Test prompts, limits and pricing before you switch.
121
+ - Anthropic's dates apply to the Claude API. Amazon Bedrock and Google-managed Vertex can use different dates, so check your platform's page for those ids.
122
+ - Models with no announced retirement show no date. Anthropic's "not sooner than" dates appear with `--all` and never fail a build.
123
+ - Retirement dates move. A data file that is months old is a risk: upgrade often, or pass `--data`.
124
+
125
+ ## Keep the data current
126
+
127
+ The data file is plain JSON, and it is the part that goes stale. Found a changed date or a missing model? Open a PR that edits `scripts/build_data.py`, link the provider page, and run it to regenerate the file.
128
+
129
+ ## Use it from Claude Code
130
+
131
+ `.claude/skills/amagi-model-sunset/SKILL.md` makes this a Claude Code skill. Ask "is my code using any retired models?" and Claude runs the scan and reports the replacements.
132
+
133
+ ## If this saved you time
134
+
135
+ A star helps others find it. If it caught a model before it broke production, tell us what it found.
136
+
137
+ More from AmagiLabs: [github.com/amagilabs](https://github.com/amagilabs).
138
+
139
+ ## License
140
+
141
+ MIT. Copyright (c) 2026 Amagisolutionsgroup, Inc. (AmagiLabs).
@@ -0,0 +1,123 @@
1
+ # amagi-model-sunset
2
+
3
+ Scans your repo for hard-coded LLM model ids, tells you which ones OpenAI, Anthropic or Google has retired or is about to retire, and suggests the replacement. Built for developers and small teams who would rather find out in CI than from a 404 in production.
4
+
5
+ ![amagi-model-sunset demo](assets/demo.gif)
6
+
7
+ ## Install
8
+
9
+ ```bash
10
+ pipx install amagi-model-sunset
11
+ ```
12
+
13
+ Python 3.9+, no dependencies. Nothing leaves your machine: it reads your files and a dated data file that ships with the package.
14
+
15
+ ## Use
16
+
17
+ Point it at a project:
18
+
19
+ ```bash
20
+ amagi-model-sunset .
21
+ ```
22
+
23
+ This run is against the demo app in the repo (`examples/demo-app`: a Python file, a YAML config and an `.env.example`). `--today` pins the date so the output below is reproducible:
24
+
25
+ ```bash
26
+ amagi-model-sunset examples/demo-app --today 2026-11-16
27
+ ```
28
+
29
+ ```text
30
+ config/llm.yaml:3 gemini-1.5-pro-002 -> RETIRED 2025-09-24 (418d ago)
31
+ suggest: gemini-3.6-flash
32
+ config/llm.yaml:4 gemini-1.5-pro-002 -> RETIRED 2025-09-24 (418d ago)
33
+ suggest: gemini-3.6-flash
34
+ config/llm.yaml:6 anthropic.claude-3-5-sonnet-20241022-v2:0 -> RETIRED 2025-10-28 (384d ago)
35
+ suggest: claude-sonnet-5-5
36
+ app.py:14 claude-3-opus-20240229 -> RETIRED 2026-01-05 (315d ago)
37
+ suggest: claude-opus-4-8
38
+ config/llm.yaml:7 us.anthropic.claude-3-opus-20240229-v1:0 -> RETIRED 2026-01-05 (315d ago)
39
+ suggest: claude-opus-4-8
40
+ .env.example:3 text-embedding-004 -> RETIRED 2026-01-14 (306d ago)
41
+ suggest: gemini-embedding-2
42
+ .env.example:2 gpt-4-turbo -> RETIRED 2026-10-23 (24d ago)
43
+ suggest: gpt-5.6-sol
44
+ app.py:6 gpt-4-0613 -> RETIRED 2026-10-23 (24d ago)
45
+ suggest: gpt-5.6-sol
46
+ config/llm.yaml:8 claude-sonnet-4-5-20250929 -> RETIRES 2026-11-30 (in 14d)
47
+ suggest: claude-sonnet-5-5
48
+ app.py:7 gpt-4o-mini -> RETIRES 2027-01-14 (in 59d)
49
+ suggest: gpt-6-luna
50
+
51
+ source: https://cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions (checked 2026-11-16)
52
+ source: https://platform.claude.com/docs/en/about-claude/model-deprecations (checked 2026-11-16)
53
+ source: https://ai.google.dev/gemini-api/docs/deprecations (checked 2026-11-16)
54
+ source: https://platform.openai.com/docs/deprecations (checked 2026-11-16)
55
+ scanned examples/demo-app: 8 model id(s) in 10 place(s): 6 retired, 2 retiring, 0 deprecated without a date, 0 ok
56
+ FAIL: 9 place(s) use a model that is retired or retires within 30 days
57
+ data checked 2026-11-16 against the providers' own pages; run with --format json for sources
58
+ ```
59
+
60
+ It finds plain ids (`gpt-4-0613`), env defaults, YAML values, Bedrock ids and ARNs (`us.anthropic.claude-3-opus-20240229-v1:0`), Vertex paths and `@date` ids, and fine-tune ids (`ft:gpt-4o-mini-...`).
61
+
62
+ ## Fail CI before the 404s start
63
+
64
+ ```bash
65
+ amagi-model-sunset . --fail-within 30
66
+ ```
67
+
68
+ Exit code 1 when a model is already retired or retires within 30 days, 0 otherwise. Use `--fail-within 0` to fail only on models that are already gone. A GitHub Actions step:
69
+
70
+ ```yaml
71
+ - run: pipx install amagi-model-sunset
72
+ - run: amagi-model-sunset . --fail-within 30 --exclude "docs/*"
73
+ ```
74
+
75
+ Dates change, so run it on a weekly schedule as well as on pull requests.
76
+
77
+ More options:
78
+
79
+ ```bash
80
+ amagi-model-sunset . --format json # machine-readable, with source URL and checked-on date
81
+ amagi-model-sunset . --format md # table for a PR comment
82
+ amagi-model-sunset . --all # also list models with no announced retirement
83
+ amagi-model-sunset . --exclude "*.md" # skip files (repeatable)
84
+ amagi-model-sunset . --list-models # every id the data file knows
85
+ amagi-model-sunset . --data my-models.json # use your own or a newer data file
86
+ ```
87
+
88
+ Add `# model-sunset: ignore` to a line to skip it (a migration note, a test that checks old ids).
89
+
90
+ ## Where the dates come from
91
+
92
+ `src/amagi_model_sunset/data/models.json` holds about 245 ids. Each entry carries the provider page it was copied from and the date it was checked (2026-11-16, or 2026-11-20 for rows added that day):
93
+
94
+ - OpenAI: https://platform.openai.com/docs/deprecations
95
+ - Anthropic: https://platform.claude.com/docs/en/about-claude/model-deprecations
96
+ - Google Gemini API: https://ai.google.dev/gemini-api/docs/deprecations
97
+ - Google Vertex AI (Gemini 1.x): https://cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions
98
+
99
+ Limits to know about:
100
+
101
+ - Entries marked `verified: false` were not confirmed on the provider page (older aliases that dropped off it). The tool labels them `[unverified date]`.
102
+ - "Suggest" is the provider's own recommendation, not a drop-in guarantee. Test prompts, limits and pricing before you switch.
103
+ - Anthropic's dates apply to the Claude API. Amazon Bedrock and Google-managed Vertex can use different dates, so check your platform's page for those ids.
104
+ - Models with no announced retirement show no date. Anthropic's "not sooner than" dates appear with `--all` and never fail a build.
105
+ - Retirement dates move. A data file that is months old is a risk: upgrade often, or pass `--data`.
106
+
107
+ ## Keep the data current
108
+
109
+ The data file is plain JSON, and it is the part that goes stale. Found a changed date or a missing model? Open a PR that edits `scripts/build_data.py`, link the provider page, and run it to regenerate the file.
110
+
111
+ ## Use it from Claude Code
112
+
113
+ `.claude/skills/amagi-model-sunset/SKILL.md` makes this a Claude Code skill. Ask "is my code using any retired models?" and Claude runs the scan and reports the replacements.
114
+
115
+ ## If this saved you time
116
+
117
+ A star helps others find it. If it caught a model before it broke production, tell us what it found.
118
+
119
+ More from AmagiLabs: [github.com/amagilabs](https://github.com/amagilabs).
120
+
121
+ ## License
122
+
123
+ MIT. Copyright (c) 2026 Amagisolutionsgroup, Inc. (AmagiLabs).
@@ -0,0 +1,35 @@
1
+ [build-system]
2
+ requires = ["setuptools>=68"]
3
+ build-backend = "setuptools.build_meta"
4
+
5
+ [project]
6
+ name = "amagi-model-sunset"
7
+ version = "0.1.0"
8
+ description = "Finds hard-coded LLM model ids in your repo and tells you which ones the provider is retiring, and when"
9
+ readme = "README.md"
10
+ requires-python = ">=3.9"
11
+ license = { text = "MIT" }
12
+ authors = [{ name = "AmagiLabs (Amagisolutionsgroup, Inc.)", email = "dev@amagisolutionsgroup.com" }]
13
+ keywords = ["llm", "openai", "anthropic", "gemini", "deprecation", "ci", "cli"]
14
+ classifiers = [
15
+ "Programming Language :: Python :: 3",
16
+ "License :: OSI Approved :: MIT License",
17
+ "Operating System :: OS Independent",
18
+ "Environment :: Console",
19
+ ]
20
+ dependencies = []
21
+
22
+ [project.optional-dependencies]
23
+ dev = ["pytest>=7.0"]
24
+
25
+ [project.scripts]
26
+ amagi-model-sunset = "amagi_model_sunset.cli:main"
27
+
28
+ [tool.setuptools.packages.find]
29
+ where = ["src"]
30
+
31
+ [tool.setuptools.package-data]
32
+ amagi_model_sunset = ["data/*.json"]
33
+
34
+ [tool.pytest.ini_options]
35
+ testpaths = ["tests"]
@@ -0,0 +1,4 @@
1
+ [egg_info]
2
+ tag_build =
3
+ tag_date = 0
4
+
@@ -0,0 +1,3 @@
1
+ """amagi-model-sunset: find LLM model ids that a provider is retiring."""
2
+
3
+ __version__ = "0.1.0"
@@ -0,0 +1,5 @@
1
+ import sys
2
+
3
+ from .cli import main
4
+
5
+ sys.exit(main())
@@ -0,0 +1,135 @@
1
+ """amagi-model-sunset command line."""
2
+ from __future__ import annotations
3
+
4
+ import argparse
5
+ import datetime as dt
6
+ import json
7
+ import os
8
+ import sys
9
+ from collections import OrderedDict
10
+ from typing import List, Optional
11
+
12
+ from . import __version__
13
+ from .core import Hit, failing, load_data, scan_path
14
+
15
+ ORDER = {"retired": 0, "retiring": 1, "unknown-date": 2, "active": 3}
16
+
17
+
18
+ def _label(h: Hit) -> str:
19
+ e = h.entry
20
+ if e.status == "retired":
21
+ return "RETIRED %s (%dd ago)" % (e.retires, -e.days_left)
22
+ if e.status == "retiring":
23
+ return "RETIRES %s (in %dd)" % (e.retires, e.days_left)
24
+ if e.status == "unknown-date":
25
+ return "DEPRECATED, no date"
26
+ return "ok"
27
+
28
+
29
+ def _sorted(hits: List[Hit]) -> List[Hit]:
30
+ return sorted(hits, key=lambda h: (ORDER[h.entry.status], h.entry.days_left if h.entry.days_left is not None else 0,
31
+ h.path, h.line))
32
+
33
+
34
+ def render_text(hits: List[Hit], show_all: bool, failed: List[Hit], within: int, scanned: str) -> str:
35
+ shown = [h for h in _sorted(hits) if show_all or h.entry.status != "active"]
36
+ out = []
37
+ for h in shown:
38
+ e = h.entry
39
+ flag = "" if e.verified else " [unverified date]"
40
+ out.append("%s:%d %s -> %s%s" % (h.path, h.line, h.text, _label(h), flag))
41
+ if e.status != "active" and e.replacement:
42
+ out.append(" suggest: %s" % e.replacement)
43
+ elif e.status == "active" and e.not_before:
44
+ out.append(" no retirement announced; provider says not sooner than %s" % e.not_before)
45
+ uniq = OrderedDict()
46
+ for h in hits:
47
+ uniq.setdefault((h.entry.status, h.model), 0)
48
+ uniq[(h.entry.status, h.model)] += 1
49
+ n = lambda s: sum(1 for (st, _m) in uniq if st == s) # noqa: E731
50
+ out.append("")
51
+ srcs = OrderedDict()
52
+ for h in _sorted(hits):
53
+ if h.entry.status != "active":
54
+ srcs.setdefault(h.entry.source, h.entry.checked)
55
+ for src, checked in srcs.items():
56
+ out.append("source: %s (checked %s)" % (src, checked))
57
+ out.append("scanned %s: %d model id(s) in %d place(s): %d retired, %d retiring, %d deprecated without a date, %d ok"
58
+ % (scanned, len(uniq), len(hits), n("retired"), n("retiring"), n("unknown-date"), n("active")))
59
+ if failed:
60
+ out.append("FAIL: %d place(s) use a model that is retired or retires within %d days" % (len(failed), within))
61
+ else:
62
+ out.append("PASS: nothing retired or retiring within %d days" % within)
63
+ return "\n".join(out)
64
+
65
+
66
+ def render_md(hits: List[Hit], show_all: bool, failed: List[Hit], within: int) -> str:
67
+ shown = [h for h in _sorted(hits) if show_all or h.entry.status != "active"]
68
+ out = ["| status | file | model | replacement | source |", "|---|---|---|---|---|"]
69
+ for h in shown:
70
+ e = h.entry
71
+ out.append("| %s | `%s:%d` | `%s` | %s | %s |" % (_label(h), h.path, h.line, h.text, e.replacement or "-", e.source))
72
+ out.append("")
73
+ out.append("%s: %d place(s) retired or retiring within %d days." % ("FAIL" if failed else "PASS", len(failed), within))
74
+ return "\n".join(out)
75
+
76
+
77
+ def build_parser() -> argparse.ArgumentParser:
78
+ p = argparse.ArgumentParser(
79
+ prog="amagi-model-sunset",
80
+ description="Find hard-coded LLM model ids in a repo and report each one's announced retirement date.",
81
+ epilog="Exit codes: 0 clean, 1 a model is retired or retires within --fail-within days, 2 usage error.",
82
+ )
83
+ p.add_argument("path", nargs="?", default=".", help="directory or file to scan (default: current directory)")
84
+ p.add_argument("--fail-within", type=int, default=30, metavar="DAYS",
85
+ help="exit 1 if a model is already retired or retires within DAYS days (default 30; 0 = only retired)")
86
+ p.add_argument("--format", choices=["text", "json", "md"], default="text", help="output format")
87
+ p.add_argument("--all", action="store_true", help="also list models with no retirement announced")
88
+ p.add_argument("--exclude", action="append", default=[], metavar="GLOB",
89
+ help="skip files matching this glob (repeatable)")
90
+ p.add_argument("--data", metavar="FILE", help="use this models.json instead of the bundled one")
91
+ p.add_argument("--today", metavar="YYYY-MM-DD", help="pretend today is this date (for CI reproducibility)")
92
+ p.add_argument("--list-models", action="store_true", help="print every model id in the data file and exit")
93
+ p.add_argument("--version", action="version", version="amagi-model-sunset " + __version__)
94
+ return p
95
+
96
+
97
+ def main(argv: Optional[List[str]] = None) -> int:
98
+ args = build_parser().parse_args(argv)
99
+ try:
100
+ today = dt.date.fromisoformat(args.today) if args.today else dt.date.today()
101
+ except ValueError:
102
+ print("error: --today must be YYYY-MM-DD", file=sys.stderr)
103
+ return 2
104
+ try:
105
+ data = load_data(args.data)
106
+ except (OSError, ValueError, KeyError) as exc:
107
+ print("error: cannot read data file: %s" % exc, file=sys.stderr)
108
+ return 2
109
+ if args.list_models:
110
+ for key in sorted(data.by_id):
111
+ rec = data.by_id[key]
112
+ print("%-45s %-10s %s" % (key, rec["provider"], rec.get("retires") or "-"))
113
+ return 0
114
+ if not os.path.exists(args.path):
115
+ print("error: %s does not exist" % args.path, file=sys.stderr)
116
+ return 2
117
+ if args.fail_within < 0:
118
+ print("error: --fail-within must be 0 or more", file=sys.stderr)
119
+ return 2
120
+ hits = scan_path(args.path, data, today, args.exclude)
121
+ failed = failing(hits, args.fail_within)
122
+ if args.format == "json":
123
+ print(json.dumps({"scanned": args.path, "today": today.isoformat(), "data_checked": data.checked,
124
+ "fail_within_days": args.fail_within, "failed": len(failed),
125
+ "hits": [h.to_dict() for h in _sorted(hits)]}, indent=2))
126
+ elif args.format == "md":
127
+ print(render_md(hits, args.all, failed, args.fail_within))
128
+ else:
129
+ print(render_text(hits, args.all, failed, args.fail_within, args.path))
130
+ print("data checked %s against the providers' own pages; run with --format json for sources" % data.checked)
131
+ return 1 if failed else 0
132
+
133
+
134
+ if __name__ == "__main__":
135
+ sys.exit(main())
@@ -0,0 +1,195 @@
1
+ """Find hard-coded LLM model ids in a source tree and look up their retirement dates."""
2
+ from __future__ import annotations
3
+
4
+ import datetime as dt
5
+ import fnmatch
6
+ import json
7
+ import os
8
+ import re
9
+ from dataclasses import dataclass, field
10
+ from importlib import resources
11
+ from typing import Dict, Iterable, Iterator, List, Optional
12
+
13
+ SKIP_DIRS = {
14
+ ".git", ".hg", ".svn", "node_modules", ".venv", "venv", "env", "__pycache__", ".tox", ".mypy_cache",
15
+ ".pytest_cache", ".ruff_cache", "dist", "build", ".next", ".idea", ".vscode", "site-packages", "target",
16
+ "vendor", ".terraform",
17
+ }
18
+ BINARY_EXT = {
19
+ ".png", ".jpg", ".jpeg", ".gif", ".webp", ".ico", ".pdf", ".zip", ".gz", ".tgz", ".bz2", ".xz", ".7z", ".whl",
20
+ ".jar", ".class", ".so", ".dll", ".exe", ".dylib", ".pyc", ".woff", ".woff2", ".ttf", ".otf", ".mp3", ".mp4",
21
+ ".mov", ".onnx", ".gguf", ".safetensors", ".bin", ".pt", ".parquet", ".sqlite", ".db", ".lock", ".svg",
22
+ }
23
+ MAX_BYTES = 1_000_000
24
+ IGNORE_MARK = "model-sunset: ignore"
25
+
26
+ # one candidate token: letters/digits plus . _ : @ - (so ids, Bedrock ids and ARN tails stay whole)
27
+ TOKEN_RE = re.compile(r"[A-Za-z0-9][A-Za-z0-9._:@-]*[A-Za-z0-9]")
28
+ REGION_PREFIX = re.compile(r"^(?:us-gov|us|eu|apac|jp|au|global)\.")
29
+ VENDOR_PREFIX = re.compile(r"^(?:anthropic|openai|google|meta|amazon)\.")
30
+ BEDROCK_SUFFIX = re.compile(r"(?:-v\d+)?(?::\d+)?$")
31
+
32
+
33
+ @dataclass
34
+ class Entry:
35
+ provider: str
36
+ id: str
37
+ status: str # retired | retiring | active | unknown-date
38
+ retires: Optional[str]
39
+ days_left: Optional[int]
40
+ replacement: str
41
+ source: str
42
+ checked: str
43
+ verified: bool
44
+ note: str = ""
45
+ not_before: Optional[str] = None
46
+
47
+
48
+ @dataclass
49
+ class Hit:
50
+ path: str
51
+ line: int
52
+ text: str # the id as written in the file
53
+ model: str # normalized id that matched the data file
54
+ entry: Entry
55
+
56
+ def to_dict(self) -> dict:
57
+ e = self.entry
58
+ return {
59
+ "path": self.path, "line": self.line, "found": self.text, "model": self.model, "provider": e.provider,
60
+ "status": e.status, "retires": e.retires, "days_left": e.days_left, "replacement": e.replacement,
61
+ "source": e.source, "checked": e.checked, "verified": e.verified, "note": e.note,
62
+ }
63
+
64
+
65
+ @dataclass
66
+ class Data:
67
+ checked: str
68
+ by_id: Dict[str, dict] = field(default_factory=dict)
69
+
70
+
71
+ def load_data(path: Optional[str] = None) -> Data:
72
+ if path:
73
+ with open(path, encoding="utf-8") as fh:
74
+ doc = json.load(fh)
75
+ else:
76
+ doc = json.loads(resources.files("amagi_model_sunset").joinpath("data/models.json").read_text(encoding="utf-8"))
77
+ data = Data(checked=doc.get("checked", ""))
78
+ for m in doc["models"]:
79
+ for mid in m["ids"]:
80
+ data.by_id[mid.lower()] = m
81
+ return data
82
+
83
+
84
+ def parse_date(s: str) -> dt.date:
85
+ return dt.date.fromisoformat(s)
86
+
87
+
88
+ def make_entry(model_id: str, rec: dict, today: dt.date) -> Entry:
89
+ retires = rec.get("retires")
90
+ if retires:
91
+ days = (parse_date(retires) - today).days
92
+ status = "retired" if days <= 0 else "retiring"
93
+ else:
94
+ days = None
95
+ status = "unknown-date" if rec.get("note", "").lower().startswith(("deprecated", "not found")) or not rec.get(
96
+ "verified", True) else "active"
97
+ return Entry(
98
+ provider=rec["provider"], id=model_id, status=status, retires=retires, days_left=days,
99
+ replacement=rec.get("replacement", ""), source=rec.get("source", ""), checked=rec.get("checked", ""),
100
+ verified=bool(rec.get("verified", True)), note=rec.get("note", ""), not_before=rec.get("not_before"),
101
+ )
102
+
103
+
104
+ def normalize(token: str) -> List[str]:
105
+ """Return candidate lookup keys for a raw token (Bedrock, Vertex, fine-tune and ARN spellings)."""
106
+ t = token.lower()
107
+ if t.startswith("ft:"):
108
+ parts = t.split(":")
109
+ t = parts[1] if len(parts) > 1 else t
110
+ keys = [t]
111
+ t2 = REGION_PREFIX.sub("", t)
112
+ t2 = VENDOR_PREFIX.sub("", t2)
113
+ keys.append(t2)
114
+ t3 = BEDROCK_SUFFIX.sub("", t2)
115
+ keys.append(t3)
116
+ for k in list(keys):
117
+ if "@" in k:
118
+ keys.append(re.sub(r"@(\d{8})$", r"-\1", k))
119
+ keys.append(k.split("@")[0])
120
+ out: List[str] = []
121
+ for k in keys:
122
+ if k and k not in out:
123
+ out.append(k)
124
+ return out
125
+
126
+
127
+ def looks_like_model(token: str) -> bool:
128
+ return any(token.lower().startswith(p) or p in token.lower() for p in (
129
+ "gpt", "claude", "gemini", "o1", "o3", "o4", "tts-", "whisper", "text-", "chatgpt", "codex", "embedding",
130
+ "imagen", "veo", "deep-research", "gemini", "ft:", "dall-e", "sora", "babbage", "davinci",
131
+ "computer-use", "antigravity"))
132
+
133
+
134
+ def scan_text(text: str, path: str, data: Data, today: dt.date) -> Iterator[Hit]:
135
+ for n, line in enumerate(text.splitlines(), 1):
136
+ if IGNORE_MARK in line:
137
+ continue
138
+ for m in TOKEN_RE.finditer(line):
139
+ raw = m.group(0)
140
+ if not looks_like_model(raw):
141
+ continue
142
+ for key in normalize(raw):
143
+ rec = data.by_id.get(key)
144
+ if rec:
145
+ yield Hit(path, n, raw, key, make_entry(key, rec, today))
146
+ break
147
+
148
+
149
+ def iter_files(root: str, excludes: Iterable[str] = ()) -> Iterator[str]:
150
+ excludes = list(excludes)
151
+ if os.path.isfile(root):
152
+ yield root
153
+ return
154
+ for dirpath, dirnames, filenames in os.walk(root):
155
+ dirnames[:] = sorted(d for d in dirnames if d not in SKIP_DIRS)
156
+ for fn in sorted(filenames):
157
+ full = os.path.join(dirpath, fn)
158
+ rel = os.path.relpath(full, root).replace(os.sep, "/")
159
+ if os.path.splitext(fn)[1].lower() in BINARY_EXT:
160
+ continue
161
+ if any(fnmatch.fnmatch(rel, pat) or fnmatch.fnmatch(fn, pat) for pat in excludes):
162
+ continue
163
+ try:
164
+ if os.path.getsize(full) > MAX_BYTES:
165
+ continue
166
+ except OSError:
167
+ continue
168
+ yield full
169
+
170
+
171
+ def scan_path(root: str, data: Data, today: dt.date, excludes: Iterable[str] = ()) -> List[Hit]:
172
+ hits: List[Hit] = []
173
+ base = root if os.path.isdir(root) else os.path.dirname(root) or "."
174
+ for full in iter_files(root, excludes):
175
+ try:
176
+ with open(full, "rb") as fh:
177
+ raw = fh.read()
178
+ except OSError:
179
+ continue
180
+ if b"\x00" in raw[:4096]:
181
+ continue
182
+ text = raw.decode("utf-8", errors="replace")
183
+ rel = os.path.relpath(full, base).replace(os.sep, "/")
184
+ hits.extend(scan_text(text, rel, data, today))
185
+ return hits
186
+
187
+
188
+ def failing(hits: List[Hit], within_days: int) -> List[Hit]:
189
+ """Hits that should fail CI: already retired, or retiring within `within_days`."""
190
+ bad = []
191
+ for h in hits:
192
+ e = h.entry
193
+ if e.status == "retired" or (e.status == "retiring" and e.days_left is not None and e.days_left <= within_days):
194
+ bad.append(h)
195
+ return bad