noesis_ai 0.0.0__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.
@@ -0,0 +1,21 @@
1
+ MIT License
2
+
3
+ Copyright (c) 2026 Mirai Engineering OÜ
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 +1,417 @@
1
+ Metadata-Version: 2.4
2
+ Name: noesis_ai
3
+ Version: 0.0.0
4
+ Summary: Thin Python SDK and local runner for Noesis application evaluation
5
+ Author: Mirai Engineering OÜ
6
+ License-Expression: MIT
7
+ Requires-Python: >=3.12
8
+ Description-Content-Type: text/markdown
9
+ License-File: LICENSE
10
+ Requires-Dist: httpx>=0.28.1
11
+ Dynamic: license-file
12
+
13
+ # Noesis SDK: thin workflow attachment and independent runner
14
+
15
+ `noesis_ai` is a Python >=3.12 network-only SDK. Imports: `noesis_ai` and
16
+ `noesis_ai_runner`; dependency: HTTPX. It does not install the hosted dashboard,
17
+ Streamlit, AgentOpt, backend, or provider SDKs. This README accompanies release
18
+ `noesis_ai==0.0.0`, licensed under MIT by Mirai Engineering OÜ. Install the
19
+ versioned package into the og app's existing environment; do not patch the SDK
20
+ or original app files.
21
+
22
+ ## Two independent offerings
23
+
24
+ **Serving:** use an ordinary OpenAI-compatible client against
25
+ `https://noesis-production-efed.up.railway.app/live/v1`, with a Noesis application key and an approved stable
26
+ role alias. No SDK, decorators or running evaluation process is required for
27
+ ordinary inference after the application's connection/assignments are configured.
28
+ Noesis resolves explicitly approved OSS assignments. Hosted frontier BYOK is retained but paused; ask Bruno before deleting or reactivating it.
29
+
30
+ **Optimization:** add a thin attachment around the actual original workflow and a
31
+ separate noesis runner process. Endpoint traces alone do not capture the complete
32
+ application outcome. The existing `Application`, `Noesis.run` and explicit HTTPX
33
+ hook provide that boundary; no decorators or workflow rewrite are needed.
34
+
35
+ ## Installation-agent contract
36
+
37
+ The installation README should briefly record inspection findings, the grounded
38
+ complete workload summary, proposed files, actual tested commands, and capture
39
+ versus evaluation limits. Put that README, the thin attachment, and
40
+ integration-specific tests in the user's chosen folder; suggest the Python-safe
41
+ `noesis_integration/` by default, but never force a rename or move an existing
42
+ file. Preserve the og app's namespace/imports and use real launch commands; do
43
+ not add `sys.path` hacks. If the folder cannot support the imports or commands,
44
+ explain why and ask where to write instead. If the user relocates it, update
45
+ commands and retest.
46
+
47
+ Use this README and shipped integration example. No Noesis internal architecture,
48
+ backend source, test fixtures, old customer integration or repository history is needed.
49
+ Inspect the og app source/dependencies/actual launch seam, never `.env` or secrets.
50
+ Explain the specific local install and new files, obtain one clear approval, then
51
+ install and add only the attachment/runner in the customer's project conventions.
52
+ Preserve original prompts, branching, parsing, UI, state and returned values.
53
+ If no existing configuration/invocation seam works, report the actual limitation;
54
+ do not invent a copied workflow, global monkey patch or SDK repair. The published
55
+ Noesis contract is fixed; adapt approved client/attachment seams, never add
56
+ customer-specific SDK/backend behavior to accommodate an app.
57
+
58
+ Install the versioned public distribution using the og app's dependency workflow:
59
+
60
+ ```sh
61
+ uv add noesis_ai==0.0.0
62
+ ```
63
+
64
+ No private repository checkout is required. For explicitly approved local release
65
+ validation, an SDK source directory is also installable (no wheel path required):
66
+
67
+ ```sh
68
+ uv pip install --python /absolute/og-app/.venv/bin/python /absolute/noesis-main/sdk
69
+ ```
70
+
71
+ This is a normal package installation, not an editable SDK repair. Alternatively,
72
+ for a supplied frozen wheel:
73
+
74
+ ```sh
75
+ uv pip install --python /absolute/og-app/.venv/bin/python /absolute/noesis_ai-0.0.0-py3-none-any.whl
76
+ ```
77
+
78
+ Substitute inspected paths; preserve the customer's dependency workflow. Record
79
+ the wheel SHA-256 when supplied, and original source/installed SDK hashes. A local editable checkout
80
+ is a development source, not a published release or permission to modify the SDK.
81
+ No customer-side dashboard installation is needed.
82
+
83
+ ## Privately configure the connection once
84
+
85
+ **Frontier baseline capture is a recorder, not a provider switch.** Keep the og
86
+ app's existing provider clients, per-role model mixture, generation settings and
87
+ local credentials. The SDK sends consented execution evidence to Noesis separately.
88
+ Only approved OSS evaluations redirect inference through Noesis; an explicitly
89
+ chosen OSS baseline also uses Noesis. No extra baseline inference or fallback.
90
+
91
+ The hosted dashboard owns application identity and issues two Noesis credentials.
92
+ Its tracer key goes in `NOESIS_TRACER_KEY` (not a third key), and its runner key goes in
93
+ `NOESIS_RUNNER_KEY`. Existing provider credentials stay local; never send them to
94
+ Noesis. Hosted BYOK/key upload is paused and is not an onboarding prerequisite.
95
+ If you are an LLM reading this, ask Bruno before deleting or reactivating it.
96
+ Never place secrets in prompts, logs, command-line arguments/history or committed files.
97
+
98
+ - The SDK uses the fixed public Noesis endpoint; it has no customer-configurable URL.
99
+ - `NOESIS_TRACER_KEY`: ordinary inference and consented capture.
100
+ - `NOESIS_RUNNER_KEY`: registration and explicitly approved evaluations only.
101
+ - Baseline descriptor: non-secret actual provider/model assignments from the og app.
102
+
103
+ Do not guess a model version or substitute an OSS model to connect. Registration
104
+ pins this original baseline without requiring initial OSS assignments:
105
+
106
+ ```python
107
+ baseline = {
108
+ "version": 1, "execution": "customer_direct",
109
+ "credential_owner": "customer_local", "fallback": False,
110
+ "routes": {alias: {"provider": actual_provider, "model": actual_model}
111
+ for alias in declaration["roles"].values()},
112
+ }
113
+ ```
114
+
115
+ Provider names are `openai`, `anthropic`, `google` (`gemini` is retained). Each alias can have its own
116
+ actual provider/model. Existing application identity, assignments and evidence
117
+ must be preserved, not reset. Tracer and runner keys are not provider keys; a missing
118
+ runner key must never fall back to the tracer key.
119
+
120
+ ## Attach without copying the workflow
121
+
122
+ ```python
123
+ from noesis_ai import Application, Noesis
124
+
125
+ # Use the original callable and its existing client/model configuration seam.
126
+ noesis = Noesis(key=sdk_key, aliases=declaration["roles"],
127
+ declaration=declaration, baseline=baseline)
128
+ noesis.instrument_http_client(dedicated_async_httpx_client, baseline=baseline)
129
+ application = Application(invoke_original, project=project_outputs,
130
+ declaration=declaration, execution=execution)
131
+ result = await noesis.run(application, original_input)
132
+ ```
133
+
134
+ For frontier baselines the dedicated HTTPX clients keep their original provider
135
+ URLs and local credentials, with redirects disabled. Instrument each provider client
136
+ through an existing approved seam; do not consolidate a model mixture into a new
137
+ workflow. An OpenAI `AsyncOpenAI` client can receive one via `http_client=...`.
138
+ The SDK's fixed URL is separately used for control/evidence; only approved trial
139
+ calls use that URL plus `/live/v1/` with runner claims and no provider credentials.
140
+ The hook links calls without buffering streamed delivery or repairing settings. Avoid automatic retry behavior
141
+ changes; retries belong to the og app and count as real attempts.
142
+
143
+ `invoke_original` is async and accepts one JSON value. It calls the original
144
+ workflow with its original preparation and logic, configuring stable aliases via
145
+ an existing seam. `project` is optional, synchronous and deterministic: select
146
+ real final values into the declared review fields, not a new evaluator or an LLM
147
+ projection. Original JSON input/output is retained separately and the original
148
+ result is returned unchanged.
149
+
150
+ The declaration is exactly `{input, roles, outputs}`: `input={"type":"text"}`;
151
+ 1–3 named roles mapped to distinct aliases sharing one prefix; 1–3 named output
152
+ fields with `kind` label/number/text. Labels may declare `config.choices`; numbers
153
+ require `config.tolerance` and optional min/max; text has no declaration config.
154
+ Structured original results are projected, not discarded. Review instructions
155
+ belong in the hosted dashboard, not a customer `eval_fn`.
156
+
157
+ Default execution is one call per role. Repeated calls require
158
+ `execution={"version":1,"coverage":{"kind":"fixed_alias_counts","counts":{alias:count}}}`,
159
+ positive counts totaling <=32. Variable/unbounded tool loops are not automatically
160
+ replayable. All model work, stream consumption and tool work must be awaited before
161
+ returning the final result. Detached work is unsupported.
162
+
163
+ Capture must run inside the actual og app execution process. Normal og-app usage
164
+ and capture are the same original workflow with explicit consent; the separate
165
+ local-evaluation runner is needed later to execute approved optimization trials
166
+ and is not a remote source upload. Wire the existing
167
+ UI/endpoint injection seam to the wrapped callable, without copying its renderer.
168
+ A separately wrapped function that the UI never calls is not integration. An
169
+ adjacent noesis runner cannot observe an untouched process.
170
+
171
+ ## Independent noesis runner
172
+
173
+ ### Short project CLI
174
+
175
+ The installation agent records a minimal, non-secret `noesis.toml` inside the
176
+ integration folder (for example `noesis_integration/noesis.toml`):
177
+
178
+ ```toml
179
+ [noesis]
180
+ entrypoint = "noesis_integration.main_with_noesis:create_runner"
181
+ # Optional existing executable attachment module, required for `noesis tracer`.
182
+ tracer_module = "noesis_integration.main_with_noesis"
183
+ # Paths are relative to this config; keep the og app root as working directory.
184
+ project_dir = ".."
185
+ # Optional private file at the og app root; loaded by uv, never parsed by the SDK.
186
+ env_file = "../.env"
187
+ ```
188
+
189
+ From the og app repository root, use `uv run noesis register` for metadata registration only,
190
+ `uv run noesis tracer` to launch your existing og-app-with-sdk attachment,
191
+ or `uv run noesis runner` only after explicit approval to execute approved jobs.
192
+ The tracer shortcut is equivalent to launching the configured module with Python;
193
+ it does not run evaluations, enable capture or grant consent. Grant consent and
194
+ click **Start tracing** in the noesis dashboard separately. Normal application
195
+ inference still uses your local provider credentials and billing. Missing or invalid
196
+ `tracer_module` fails before launch; existing register/runner configs need no change.
197
+ An activated project environment also supports `noesis register` / `noesis tracer` / `noesis runner`. The
198
+ CLI finds a single `noesis.toml` in the current directory or an immediate child
199
+ folder; it does not recursively search or climb parent directories. Multiple
200
+ configs fail closed; use `--config PATH` for an explicit or deeper location.
201
+ Bad configuration fails closed without printing values. An omitted `env_file`
202
+ uses the inherited environment or the factory's approved loader. When present,
203
+ its dotenv loading is delegated to the existing `uv run --env-file` mechanism;
204
+ the SDK does not parse dotenv files or expose secrets. The recorded config
205
+ contains an optional relative `project_dir` determining the factory's working
206
+ directory (default: config directory, preserving existing root-level configs).
207
+ Both `project_dir` and `env_file` resolve relative to the config. The installed interpreter
208
+ runs the existing module runner without dependency syncing or recursive CLI startup.
209
+ The legacy
210
+ `python -m noesis_ai_runner module:factory --register-only` command remains
211
+ supported.
212
+
213
+
214
+ The attachment exposes an async context-manager factory, e.g. `create_runner`,
215
+ yielding `(Noesis, Application)` and owning both HTTP-client lifetimes. It binds
216
+ the same original callable, declaration/projection and execution counts, using
217
+ `NOESIS_RUNNER_KEY` instead of the tracer key.
218
+
219
+ The dashboard creates the application and both one-time keys first. Installation
220
+ explains that the tracer key records consented activity while the runner key
221
+ connects configuration and approved local evaluations; provider keys stay local.
222
+ It asks for the og app repository root, a user-chosen integration-folder location
223
+ (suggesting the Python-safe `<repo-root>/noesis_integration/`), and the user's
224
+ preferred environment loader.
225
+ Default: private Git-ignored `.env`, explicitly loaded at each launch; no repeated
226
+ exports. Never read or print credential values, overwrite existing private files,
227
+ or put keys in chat, command arguments or version control.
228
+
229
+ After the user approves the installation proposal and offline checks pass, run
230
+ metadata registration only, from the actual attachment's importable directory:
231
+
232
+ ```sh
233
+ uv run --env-file .env python -m noesis_ai_runner package.attachment:create_runner --register-only
234
+ ```
235
+
236
+ Replace `.env` loading with the customer's existing mechanism if preferred. The
237
+ command requires an accepted `ready_for_capture=true` receipt, closes its factory,
238
+ and exits without polling, workflow invocation or inference. A rejected/missing
239
+ readiness receipt means setup is incomplete, even if registration was accepted.
240
+ The factory must not execute work on import or entry. Registration sends the
241
+ actual declaration/baseline metadata, not provider keys or application source.
242
+
243
+ Only later, after explicit evaluation-execution approval:
244
+
245
+ ```sh
246
+ uv run --env-file .env python -m noesis_ai_runner package.attachment:create_runner
247
+ ```
248
+
249
+ The runner registers the unchanged declaration/baseline before polling; repeated
250
+ registration is idempotent. It invokes the original workflow under temporary
251
+ server-approved OSS assignments. Starting polling can execute previously approved
252
+ jobs: obtain approval and isolate state, tools, writes, caches, billing side
253
+ effects and external submissions first. No automatic replay after a claimed
254
+ execution or ambiguous result delivery. Stop with Ctrl+C. AgentOpt and approved judge
255
+ judging stay server-side; optimization never activates an assignment.
256
+
257
+ Claimed result delivery requires a successful HTTP response with `accepted: true`.
258
+ A recognized `409` with the top-level code `execution_evidence_rejected` permits
259
+ one same-claim, metadata-only `RunnerResultRejected` infrastructure-error report;
260
+ then the runner stops. This does not resend outputs or execute the app again.
261
+ Unknown/uncoded rejections, invalid acknowledgements and ambiguous network errors
262
+ stop without this fallback. Human-readable error wording never controls recovery.
263
+ Deploy the compatible backend before updating the local SDK/runner; an older
264
+ backend's uncoded rejection stops the new runner but may still reach the backend
265
+ job deadline. A backend deployment alone does not update an installed SDK.
266
+
267
+ ## Protocol and evidence limits
268
+
269
+ The public interface follows OpenAI Chat Completions, not every API in
270
+ https://developers.openai.com/api/reference/overview . Supported protocol paths
271
+ are normal JSON and optional SSE, function tools/results, supported JSON/JSON
272
+ Schema output, and text/image input with text/tool output. Native Anthropic or
273
+ Google client protocols are not accepted just by changing their base URL. Noesis
274
+ owns upstream adaptation; unsupported model/feature combinations fail explicitly.
275
+ Tool execution remains in the og app. No silent parameter removal or provider
276
+ substitution. Model/account/deployment features require actual provider support.
277
+
278
+ The backend retains one canonical trace per call, incremental tool/content deltas,
279
+ finish reasons, usage and settings. Missing usage/prices remain unknown. Refusal,
280
+ truncation, malformed streams, incomplete tools and cancellation are not valid
281
+ quality evidence. Image serving/capture does not establish multimodal quality:
282
+ the text-only judge rejects image-bearing text evaluations. Audio/video, file
283
+ uploads, generated image/audio, Responses and Realtime are outside this release.
284
+
285
+ Application capture is opt-in and bounded: finite JSON, <=100 KB per SDK value,
286
+ <=10 KB display input and <=2 MB backend streamed evidence. Unsupported or oversized
287
+ capture must not become invented evidence. Capture failures warn/abort enrollment
288
+ without substituting the served answer; evaluation failures fail closed. Clients
289
+ must disable redirects to prevent an authenticated inference request escaping the
290
+ validated destination. This is backend proxy routing, not an HTTP redirect.
291
+
292
+ Direct observations are customer-SDK evidence, not server-attested inference.
293
+ Consume streams and await all model/tool work before the original workflow returns.
294
+ Incomplete observation invalidates capture, not the original served result. Usage
295
+ is observed; frontier dollar cost stays unknown without trustworthy rate provenance.
296
+ Native provider protocols and arbitrary HTTP destinations are not supported just by
297
+ changing a URL.
298
+
299
+ ## Versioned whole-outcome review
300
+
301
+ A Phase 1 declaration may use `version: 2`, JSON input, and an explicit outcome view of
302
+ `application_output` or deterministic `outputs`. The original application input/output remain
303
+ bounded authoritative evidence; the display excerpt is never replayed. Reviews may contain
304
+ nested objects, lists, scalars, booleans, and null, with optional bounded additional instructions.
305
+ The backend stores one versioned `outcome` specification and sends one approved judge request
306
+ containing the original input, reviewed reference, selected view, and complete candidate outcome.
307
+ Legacy flat declarations and field scoring retain their existing semantics. Legacy execution
308
+ version 1 retains fixed alias counts totaling at most 32 calls. New whole-outcome declarations
309
+ have application-specific roles without the former three-role ceiling. For loops, branches,
310
+ skipped roles and app-owned retries, explicitly configure approved execution bounds:
311
+
312
+ ```python
313
+ execution = {"version": 2,
314
+ "coverage": {"kind": "bounded_attempts", "max_attempts": 40},
315
+ "max_wall_seconds": 120, "max_output_tokens": 4096}
316
+ app = Application(invoke, declaration=declaration, execution=execution)
317
+ ```
318
+
319
+ These are illustrative allowances, not defaults or a dollar budget. Actual attempts can be
320
+ zero or vary between baseline and candidate paths. Every admitted attempt must reconcile
321
+ with terminal evidence before successful closure; recovered failures remain attributable,
322
+ and missing usage remains unknown cost. Output bounds reject, never silently clamp requests.
323
+ Policy integer bounds must be browser-safe JSON integers; workflow execution totals use exact
324
+ arithmetic and decimal text when beyond that range. The global 100-execution ceiling is gone,
325
+ but the exact full-comparison count still requires approval. A larger role space grows
326
+ Cartesian comparison cost; lifting a ceiling does not solve that. Leveraging AgentOpt's
327
+ existing efficient selection capabilities is later work, not a search change in this phase.
328
+ Use this updated checkout's SDK for the new contract; SDK extraction/publication is separate.
329
+
330
+ Preview checks the complete baseline/reference judge messages against the declared context
331
+ using a conservative UTF-8 byte bound, including chat framing and output reserve; unknown
332
+ limits block whole-outcome optimization. Each actual candidate payload is checked again before
333
+ judging. Nothing is silently truncated, but conservative checks can reject a payload that an
334
+ exact tokenizer could fit, and a candidate can exceed context despite its baseline fitting.
335
+ The dashboard fails closed for integers outside JavaScript's safe range rather than silently
336
+ rounding reviewed references. Original bounded JSON remains preserved by the backend.
337
+
338
+ ## Acceptance and honest handoff
339
+
340
+ First verify offline with synthetic inputs/transports: installed module resolution,
341
+ projection, original-result preservation, invocation isolation and cleanup. Record
342
+ exact working directories and real environment-loading capture/runner commands,
343
+ not final placeholders. Verify registration/readiness before declaring installation
344
+ complete. First explain that a new, independent attachment file stays on the user's
345
+ machine and connects the dashboard by sending configuration metadata, not original
346
+ application source or provider keys. Later consented capture transmits execution
347
+ evidence; the separate evaluation runner executes the original application locally.
348
+ Use two clear handoff sections: “Start using and tracing your application now” and
349
+ “Later: run optimization experiments”. Explain each process before its command.
350
+ Then say “Open a new terminal”, give the actual `cd` and capture command, and point
351
+ to the information folder for details. The evaluation runner is not needed for
352
+ normal usage or baseline tracing and must not start until separately approved. State offline
353
+ wiring, registration readiness and live capture separately. If live capture was
354
+ not separately authorized, say so; offer one bounded synthetic-input smoke test
355
+ with explicit consent and customer-owned provider billing. Diagnose application,
356
+ provider and capture failures separately; never rewrite customer prompts/parsers
357
+ to make setup appear successful.
358
+ Quality examples and review are required before optimization, not a prerequisite
359
+ for local installation. Registration, baseline configured, capture consent, actual
360
+ complete captured executions and runner polling are different facts.
361
+
362
+ Paid baseline/candidate/judge inference, real data upload, hosted runner polling,
363
+ deployment and serving activation require separate explicit approval. Synthetic
364
+ checks prove wiring, not live provider compatibility, maintained quality or savings.
365
+ The existing landing-page installation prompt is the customer entrypoint; this
366
+ README and shipped examples are its SDK-side documentation, not another setup flow.
367
+
368
+ Inspection approval does not authorize Git staging, untracking a private file,
369
+ changing the index, editing `.gitignore`, or rewriting history. Report tracked
370
+ private configuration without reading its values and obtain specific approval for
371
+ any such Git changes. Never automatically untrack `.env` as a security cleanup.
372
+
373
+ ## Updating an existing integration
374
+
375
+ Keep the full initial installation prompt for the first connection. For later
376
+ application changes, the shared maintenance skill is bundled at
377
+ `skills/noesis-update-integration/SKILL.md` in this SDK checkout and installed at
378
+ `<python-prefix>/share/noesis_ai/skills/noesis-update-integration/SKILL.md`.
379
+ The installation agent proposes and copies it into the current coding agent's
380
+ appropriate project-local skill directory, without overwriting existing rules or
381
+ granting permissions. If native discovery is unavailable, it records a readable
382
+ local copy and an explicit-read prompt instead. No separate per-agent SDK is needed.
383
+
384
+ Future skills repository: `<NOESIS_SKILLS_REPOSITORY_URL>` — **placeholder only**;
385
+ do not fetch it or guess a replacement. Use the approved bundled SDK snapshot now.
386
+ The integration README must record the actual skill source/version/destination,
387
+ verify the copied contents, and give a copyable update prompt with real paths.
388
+
389
+ Use the update skill when LLM calls, prompts, parsing, branching, aliases, baseline
390
+ configuration or SDK compatibility change. The agent inspects the relevant og app
391
+ context, proposes minimal integration edits and tests, and explains registration,
392
+ capture/review and evaluation requirements. It does not repeat account/key setup.
393
+
394
+ Initial integrations set `configuration_revision=1` on their capture and runner
395
+ `Noesis(...)` clients. Material approved workflow, alias, baseline or execution-policy
396
+ changes use exactly the next revision; the runner registers `Application.execution`.
397
+ Identical registration remains idempotent, including during active capture. A new
398
+ revision requires capture/operations/claims to be idle, preserves old configuration
399
+ and evidence, clears only the active-dataset selection, and requires fresh
400
+ capture/review before optimization. Same-label prompt/workflow changes still need
401
+ this deliberate update; there is no automatic application-source change detector.
402
+
403
+ Updates conflicting with saved serving labels/baseline are rejected rather than
404
+ changing production routing. Legacy backends may not support revision updates;
405
+ the skill must explain that blocker rather than bypass locks. Registration
406
+ acceptance alone does not establish live capture or validate old optimization
407
+ results for a changed workflow. Credentials, capture, polling, inference and
408
+ serving remain separately approved; verify CLI/backend versions before use.
409
+
410
+ ## Release verification
411
+
412
+ For a reproducible installation, retain the resolved `noesis_ai` version and
413
+ artifact hashes in the og app's dependency lock. The installed modules are
414
+ `noesis_ai` and `noesis_ai_runner`; the console command is `noesis`.
415
+ Do not install the unrelated `noesis` distribution. Package installation and
416
+ imports never register an application, start polling, run inference or activate
417
+ serving. Those operations retain their separate approval requirements.