LOC Client

Synchronous Client

LOC

Synchronous AIME LOC client.

Parameters:

Name	Type	Description	Default
api_key	str | None	Your AIME API key (sk-aime-{tier}_...). Falls back to the AIME_API_KEY environment variable.	None
base_url	str | None	Override the API base URL (for testing / self-hosted).	None
timeout	Timeout | None	Custom :class:httpx.Timeout.	None
max_retries	int	Number of automatic retries on transient failures.	3
poll_interval	float	Seconds between scan status polls.	2.0
poll_timeout	float	Max seconds to wait for a scan to complete.	600.0

Example::

with LOC(api_key="sk-aime-academic_...") as loc:
    profile = loc.scan("meta-llama/Llama-4-Scout")
    profile.radar_chart()

__init__(api_key=None, base_url=None, timeout=None, max_retries=3, poll_interval=2.0, poll_timeout=600.0)

scan(model, *, questions='26q', functions=None, cache=True)

Scan an AI model's cognitive profile.

Submits a scan job to the AIME Cloud, polls until completion, and returns a fully-populated :class:CognitiveProfile.

Parameters:

Name	Type	Description	Default
model	str	HuggingFace model ID (e.g., "meta-llama/Llama-4-Scout").	required
questions	str	Question set — "26q" (quick) or "78q" (full).	'26q'
functions	list[str] | None	Optional subset of cognitive functions to evaluate.	None
cache	bool	Whether to return cached results if available.	True

Returns:

Type	Description
CognitiveProfile	class:CognitiveProfile with 13-function TC scores, gate
CognitiveProfile	diagnostics, and visualization/export methods.

Example::

profile = loc.scan("meta-llama/Llama-4-Scout")
print(profile.tc_score)   # 14.2
profile.radar_chart()

compare(model_a, model_b, *, questions='26q')

Compare cognitive profiles of two models.

Parameters:

Name	Type	Description	Default
model_a	str	First model HuggingFace ID.	required
model_b	str	Second model HuggingFace ID.	required
questions	str	Question set — "26q" or "78q".	'26q'

Returns:

Type	Description
ModelComparison	class:ModelComparison with per-function deltas, winner,
ModelComparison	and visualization methods.

Example::

comp = loc.compare("Llama-4-Scout", "Llama-3.3-70B")
print(comp.summary())
comp.delta_chart()

training_audit(base, trained, *, method='unknown', questions='78q')

Audit what training did to cognitive coherence.

Parameters:

Name	Type	Description	Default
base	str	HuggingFace ID of the base/pretrained model.	required
trained	str	HuggingFace ID of the fine-tuned variant.	required
method	str	Training method ("SFT", "RLHF", "DPO", etc.).	'unknown'
questions	str	Question set — "26q" or "78q".	'78q'

Returns:

Type	Description
TrainingAudit	class:TrainingAudit with before/after profiles, gate changes,
TrainingAudit	and actionable recommendations.

Example::

audit = loc.training_audit(
    base="mistralai/Mistral-7B-v0.3",
    trained="mistralai/Mistral-7B-Instruct-v0.3",
    method="SFT",
)
audit.save_report("sft_audit.md")

benchmark(models, *, questions='26q')

Benchmark multiple models and generate leaderboard.

Parameters:

Name	Type	Description	Default
models	list[str]	List of HuggingFace model IDs to benchmark.	required
questions	str	Question set — "26q" or "78q".	'26q'

Returns:

Type	Description
BenchmarkResult	class:BenchmarkResult with ranked models and heatmap.

Example::

results = loc.benchmark([
    "meta-llama/Llama-4-Scout",
    "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
])
results.leaderboard_table()

leaderboard(*, top_n=20)

Get the public LOC leaderboard.

Parameters:

Name	Type	Description	Default
top_n	int	Number of top models to return.	20

Returns:

Type	Description
Leaderboard	class:Leaderboard with ranked entries.

models()

List all available/cached model IDs.

usage()

Check API usage for current billing period.

Returns:

Type	Description
dict[str, Any]	Dictionary with scans_used, scans_limit, tier, period_end, etc.

close()

Close the HTTP connection pool.

Asynchronous Client

AsyncLOC

Asynchronous AIME LOC client.

Identical API to :class:LOC but all methods are async.

Example::

async with AsyncLOC(api_key="sk-aime-academic_...") as loc:
    profile = await loc.scan("meta-llama/Llama-4-Scout")
    profile.radar_chart()

__init__(api_key=None, base_url=None, timeout=None, max_retries=3, poll_interval=2.0, poll_timeout=600.0)

scan(model, *, questions='26q', functions=None, cache=True) async

Scan an AI model's cognitive profile (async).

compare(model_a, model_b, *, questions='26q') async

Compare cognitive profiles of two models (async).

training_audit(base, trained, *, method='unknown', questions='78q') async

Audit what training did to cognitive coherence (async).

benchmark(models, *, questions='26q') async

Benchmark multiple models and generate leaderboard (async).

leaderboard(*, top_n=20) async

Get the public LOC leaderboard (async).

models() async

List all available/cached model IDs (async).

usage() async

Check API usage (async).

close() async

Close the HTTP connection pool.