laga 0.1.3

Repair malformed JSON from language models and hand-written config.

laga

laga repairs malformed JSON and returns Python objects.

It is built for the real world: model output, hand-written config, copied snippets, and other almost-JSON text that is usually close enough to recover but not strict enough for json.loads.

Highlights

• Repairs common JSON mistakes such as trailing commas, missing commas, single quotes, unquoted keys, comments, fences, smart quotes, and truncated containers.
• Tries the standard library first, so valid JSON stays fast and strict.
• Keeps the public API small: repair, repair_to_str, loads, and LagaError.
• Ships with a CLI, type information, tests, docs, and CI.

Supported Versions

laga 0.1.3 supports Python 3.10 through 3.13.

Install

For normal use from PyPI:

python -m pip install laga

For local development from a checkout:

python -m pip install -e ".[dev]"

Quick Start

import laga

text = "{name: 'Ada', active: True, roles: ['admin', 'writer',],}"

data = laga.repair(text)
print(data)
print(laga.repair_to_str(text))

CLI

You can also use laga from the command line:

laga "{name: 'Ada', active: True, roles: ['admin',]}"
laga --strict '{"a": 1,}'

Use --strict when you want to reject ambiguous repairs instead of filling in gaps. Use --pretty to format output for humans, --stdin to read from standard input, and --duplicate-keys error when duplicate keys should fail.

laga --stdin --pretty
laga --max-depth 64 --max-input-size 10000 --duplicate-keys error '{"a": 1}'

API

Function	Returns	Description
laga.repair(text, strict=False)	Python object	Repairs malformed JSON and returns Python data.
laga.repair_to_str(text, strict=False)	str	Repairs malformed JSON and returns normalized JSON text.
laga.loads(text, strict=False)	Python object	Alias for laga.repair.
laga.LagaError	Exception type	Raised when repair fails or input is unrecoverable.

What It Repairs

Rule	Before	After
Trailing comma	{"a": 1,}	{"a":1}
Missing comma	{"a": 1 "b": 2}	{"a":1,"b":2}
Single quotes	{'a': 'b'}	{"a":"b"}
Unquoted keys	{name: "Ada"}	{"name":"Ada"}
Python literals	{"ok": True, "none": None}	{"ok":true,"none":null}
Comments	{"a": 1 // note\n}	{"a":1}
Markdown fences	json\n{"a":1}\n	{"a":1}
Smart quotes	“a”	"a"
Prose around JSON	Result: {"a":1}	{"a":1}
Truncated structure	{"a": [1, 2	{"a":[1,2]}

Strict Mode

strict=True is for validation workflows where you want laga to fail instead of guessing.

In strict mode, laga rejects ambiguous recoveries such as:

• missing commas between object members or array items
• unterminated objects or arrays
• other repairs that would require filling in structure the input did not clearly provide

Behavior Notes

• json.loads is still the best choice when you control the producer and the input is already valid JSON.
• Duplicate keys use last-value-wins behavior, matching Python dict semantics.
• Use duplicate_keys="error" when you want duplicate object members to fail instead of being merged.
• laga does not evaluate expressions or try to invent arbitrary JavaScript syntax.
• The parser is intentionally conservative around malformed numbers and highly ambiguous text.

When To Use laga

Use laga when the input is noisy, user-facing, or generated by a model and you need a predictable repair step before parsing.

Use json.loads when correctness and strictness matter more than recovery.

Development

make install
make lint
make typecheck
make test
make build

Project Links

• Documentation: docs/index.md
• API details: docs/api.md
• Usage examples: docs/usage.md
• Contributing guide: CONTRIBUTING.md

Citation

If you use laga in academic work, research notes, or other formal writing, please cite the repository using the metadata in CITATION.cff.

Suggested citation format:

Laitinen-Fredriksson Lundström-Imanov, G. O. Y. (2026). laga (Version 0.1.3) [Computer software]. https://github.com/olaflaitinen/laga