artl-memory 0.2.0__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 +1,201 @@
1
+ Apache License
2
+ Version 2.0, January 2004
3
+ http://www.apache.org/licenses/
4
+
5
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
6
+
7
+ 1. Definitions.
8
+
9
+ "License" shall mean the terms and conditions for use, reproduction,
10
+ and distribution as defined by Sections 1 through 9 of this document.
11
+
12
+ "Licensor" shall mean the copyright owner or entity authorized by
13
+ the copyright owner that is granting the License.
14
+
15
+ "Legal Entity" shall mean the union of the acting entity and all
16
+ other entities that control, are controlled by, or are under common
17
+ control with that entity. For the purposes of this definition,
18
+ "control" means (i) the power, direct or indirect, to cause the
19
+ direction or management of such entity, whether by contract or
20
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
21
+ outstanding shares, or (iii) beneficial ownership of such entity.
22
+
23
+ "You" (or "Your") shall mean an individual or Legal Entity
24
+ exercising permissions granted by this License.
25
+
26
+ "Source" form shall mean the preferred form for making modifications,
27
+ including but not limited to software source code, documentation
28
+ source, and configuration files.
29
+
30
+ "Object" form shall mean any form resulting from mechanical
31
+ transformation or translation of a Source form, including but
32
+ not limited to compiled object code, generated documentation,
33
+ and conversions to other media types.
34
+
35
+ "Work" shall mean the work of authorship, whether in Source or
36
+ Object form, made available under the License, as indicated by a
37
+ copyright notice that is included in or attached to the work
38
+ (an example is provided in the Appendix below).
39
+
40
+ "Derivative Works" shall mean any work, whether in Source or Object
41
+ form, that is based on (or derived from) the Work and for which the
42
+ editorial revisions, annotations, elaborations, or other modifications
43
+ represent, as a whole, an original work of authorship. For the purposes
44
+ of this License, Derivative Works shall not include works that remain
45
+ separable from, or merely link (or bind by name) to the interfaces of,
46
+ the Work and Derivative Works thereof.
47
+
48
+ "Contribution" shall mean any work of authorship, including
49
+ the original version of the Work and any modifications or additions
50
+ to that Work or Derivative Works thereof, that is intentionally
51
+ submitted to Licensor for inclusion in the Work by the copyright owner
52
+ or by an individual or Legal Entity authorized to submit on behalf of
53
+ the copyright owner. For the purposes of this definition, "submitted"
54
+ means any form of electronic, verbal, or written communication sent
55
+ to the Licensor or its representatives, including but not limited to
56
+ communication on electronic mailing lists, source code control systems,
57
+ and issue tracking systems that are managed by, or on behalf of, the
58
+ Licensor for the purpose of discussing and improving the Work, but
59
+ excluding communication that is conspicuously marked or otherwise
60
+ designated in writing by the copyright owner as "Not a Contribution."
61
+
62
+ "Contributor" shall mean Licensor and any individual or Legal Entity
63
+ on behalf of whom a Contribution has been received by Licensor and
64
+ subsequently incorporated within the Work.
65
+
66
+ 2. Grant of Copyright License. Subject to the terms and conditions of
67
+ this License, each Contributor hereby grants to You a perpetual,
68
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
69
+ copyright license to reproduce, prepare Derivative Works of,
70
+ publicly display, publicly perform, sublicense, and distribute the
71
+ Work and such Derivative Works in Source or Object form.
72
+
73
+ 3. Grant of Patent License. Subject to the terms and conditions of
74
+ this License, each Contributor hereby grants to You a perpetual,
75
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
76
+ (except as stated in this section) patent license to make, have made,
77
+ use, offer to sell, sell, import, and otherwise transfer the Work,
78
+ where such license applies only to those patent claims licensable
79
+ by such Contributor that are necessarily infringed by their
80
+ Contribution(s) alone or by combination of their Contribution(s)
81
+ with the Work to which such Contribution(s) was submitted. If You
82
+ institute patent litigation against any entity (including a
83
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
84
+ or any Contribution incorporated within the Work constitutes direct
85
+ or contributory patent infringement, then any patent licenses
86
+ granted to You under this License for that Work shall terminate
87
+ as of the date such litigation is filed.
88
+
89
+ 4. Redistribution. You may reproduce and distribute copies of the
90
+ Work or Derivative Works thereof in any medium, with or without
91
+ modifications, and in Source or Object form, provided that You
92
+ meet the following conditions:
93
+
94
+ (a) You must give any other recipients of the Work or
95
+ Derivative Works a copy of this License; and
96
+
97
+ (b) You must cause any modified files to carry prominent notices
98
+ stating that You changed the files; and
99
+
100
+ (c) You must retain, in the Source form of any Derivative Works
101
+ that You distribute, all copyright, patent, trademark, and
102
+ attribution notices from the Source form of the Work,
103
+ excluding those notices that do not pertain to any part of
104
+ the Derivative Works; and
105
+
106
+ (d) If the Work includes a "NOTICE" text file as part of its
107
+ distribution, then any Derivative Works that You distribute must
108
+ include a readable copy of the attribution notices contained
109
+ within such NOTICE file, excluding those notices that do not
110
+ pertain to any part of the Derivative Works, in at least one
111
+ of the following places: within a NOTICE text file distributed
112
+ as part of the Derivative Works; within the Source form or
113
+ documentation, if provided along with the Derivative Works; or,
114
+ within a display generated by the Derivative Works, if and
115
+ wherever such third-party notices normally appear. The contents
116
+ of the NOTICE file are for informational purposes only and
117
+ do not modify the License. You may add Your own attribution
118
+ notices within Derivative Works that You distribute, alongside
119
+ or as an addendum to the NOTICE text from the Work, provided
120
+ that such additional attribution notices cannot be construed
121
+ as modifying the License.
122
+
123
+ You may add Your own copyright statement to Your modifications and
124
+ may provide additional or different license terms and conditions
125
+ for use, reproduction, or distribution of Your modifications, or
126
+ for any Derivative Works regarding such redistribution as long as
127
+ such use, reproduction, and distribution of the Work is permitted
128
+ by the Licensor and is not in violation of the terms of this License.
129
+
130
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
131
+ any Contribution intentionally submitted for inclusion in the Work
132
+ by You to the Licensor shall be under the terms and conditions of this
133
+ License, without any additional terms or conditions.
134
+ Notwithstanding the above, nothing herein shall supersede or modify
135
+ the terms of any separate license agreement you may have executed
136
+ with Licensor regarding such Contributions.
137
+
138
+ 6. Trademarks. This License does not grant permission to use the trade
139
+ names, trademarks, service marks, or product names of the Licensor,
140
+ except as required for reasonable and customary use in describing the
141
+ origin of the Work and reproducing the origin of the Work.
142
+
143
+ 7. Disclaimer of Warranty. Unless required by applicable law or
144
+ agreed to in writing, Licensor provides the Work (and each
145
+ Contributor provides its Contributions) on an "AS IS" BASIS,
146
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
147
+ implied, including, without limitation, any warranties or conditions
148
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
149
+ PARTICULAR PURPOSE. You are solely responsible for determining the
150
+ appropriateness of using or redistributing the Work and assume any
151
+ risks associated with Your use of the Licensor's permissions.
152
+
153
+ 8. Limitation of Liability. In no event and under no legal theory,
154
+ whether in tort (including negligence), contract, or otherwise,
155
+ unless required by applicable law (such as copyright) or agreed
156
+ to in writing, will any Contributor be liable to You for damages,
157
+ including any direct, indirect, special, incidental, or consequential
158
+ damages of any character arising as a result of this License or out
159
+ of the use or inability to use the Work (including but not limited
160
+ to damages for personal injury or loss of goodwill, work stoppage,
161
+ computer failure or malfunction, or any and all other commercial
162
+ damages or losses), even if such Contributor has been advised of the
163
+ possibility of such damages.
164
+
165
+ 9. Accepting Warranty or Additional Liability. While redistributing
166
+ the Work or Derivative Works thereof, You may choose to offer,
167
+ and charge a fee for, acceptance of support, warranty, indemnity,
168
+ or other liability obligations and/or rights consistent with this
169
+ License. However, in accepting such obligations, You may act only
170
+ on behalf of You and on Your sole responsibility, not on behalf of
171
+ any other Contributor, and only if You agree to indemnify, defend,
172
+ and hold each Contributor harmless for any liability incurred by,
173
+ or claims asserted against, such Contributor by reason of your
174
+ accepting any such warranty or additional liability.
175
+
176
+ END OF TERMS AND CONDITIONS
177
+
178
+ APPENDIX: How to apply the Apache License to your work.
179
+
180
+ To apply the Apache License to your work, attach the following
181
+ boilerplate notice, with the fields enclosed by brackets "[]"
182
+ replaced with your own identifying information. (Don't include
183
+ the brackets!) The text should be enclosed in the appropriate
184
+ comment syntax for the file format. We also recommend that a
185
+ file or class name and description of purpose be included on the
186
+ same "printed page" as the copyright notice for easier
187
+ identification within third-party archives.
188
+
189
+ Copyright [yyyy] [name of copyright owner]
190
+
191
+ Licensed under the Apache License, Version 2.0 (the "License");
192
+ you may not use this file except in compliance with the License.
193
+ You may obtain a copy of the License at
194
+
195
+ http://www.apache.org/licenses/LICENSE-2.0
196
+
197
+ Unless required by applicable law or agreed to in writing, software
198
+ distributed under the License is distributed on an "AS IS" BASIS,
199
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
200
+ See the License for the specific language governing permissions and
201
+ limitations under the License.
@@ -0,0 +1,176 @@
1
+ Metadata-Version: 2.4
2
+ Name: artl-memory
3
+ Version: 0.2.0
4
+ Summary: The multi-person memory layer for agents: verbatim utterances, per-person isolation, evidence chain, confidence. It remembers everyone in the room — separately.
5
+ Author: ARTL family
6
+ License: Apache-2.0
7
+ Project-URL: Homepage, https://github.com/ARTL-family/artl-memory
8
+ Keywords: agent,memory,multi-person,evidence-chain,local-first,sqlite
9
+ Classifier: Development Status :: 3 - Alpha
10
+ Classifier: Intended Audience :: Developers
11
+ Classifier: License :: OSI Approved :: Apache Software License
12
+ Classifier: Programming Language :: Python :: 3
13
+ Classifier: Topic :: Software Development :: Libraries
14
+ Requires-Python: >=3.11
15
+ Description-Content-Type: text/markdown
16
+ License-File: LICENSE
17
+ Dynamic: license-file
18
+
19
+ # artl-memory
20
+
21
+ **The multi-person memory layer for agents.**
22
+ Verbatim utterances · per-person isolation · evidence chains · confidence scores.
23
+
24
+ > It remembers everyone in the room — separately.
25
+ > We don't promise understanding. We promise not to forget.
26
+
27
+ [中文文档](README.zh-CN.md)
28
+
29
+ artl-memory is extracted from a production assistant. It does **not** "summarize a
30
+ profile of you and pretend to know you" — it stores **who said exactly what, and when**,
31
+ each memory carrying its source utterance and a confidence score. When a fact is
32
+ corrected, the old entry is marked expired — never silently overwritten — and the
33
+ evidence chain stays in place.
34
+
35
+ > **Scope: the ledger, not the ears.** artl-memory contains no audio, no ASR, no
36
+ > voiceprint. Your upstream pipeline (diarization, voiceprint, manual labels)
37
+ > decides *who* said what and hands this library the name — it keeps the books.
38
+ > That is what makes it dependency-free: usable without ears, plug-and-play with
39
+ > them.
40
+
41
+ ```bash
42
+ pip install . # from source (PyPI package not yet published)
43
+ ```
44
+
45
+ ## Which memory layer should you pick?
46
+
47
+ Everyone arriving here has Mem0 (~60k★) as their reference point, so let's be direct
48
+ about it. These four tools answer **different questions** — this table is for picking
49
+ the right one, not for ranking them.
50
+
51
+ | | **artl-memory** | **Mem0** | **memoripy v4** | **MCP memory (official)** |
52
+ |---|---|---|---|---|
53
+ | Needs an LLM to write memory | **No** — zero model calls | Yes — extraction LLM on every `add()` | No required — deterministic extractor, bring-your-own optional | No — the agent itself decides what to store |
54
+ | What gets stored | **Verbatim utterances** + derived facts | LLM-extracted summaries | Evidence + derived memory, versioned | Entities / relations / observations |
55
+ | Can you trace a memory to its source? | Yes — `source_event` + confidence on every entry | Partial — extraction rewrites wording | Yes — citations and receipts | No |
56
+ | Multi-person isolation | **First-class** — per-person ledger, skills get scoped `MemView`s | Per-`user_id` scoping | Per-`user_id` scoping | No — one shared graph |
57
+ | Deletion | **Hard delete** — `secure_delete` + `VACUUM` + WAL truncation, verified by byte-level scan | API delete; vector-store semantics | `forget` tooling | Delete tools on the JSONL file |
58
+ | Works fully offline | Yes | No (LLM + vector store calls) | Yes | Yes |
59
+ | Semantic (vector) search | **No** — explainable bigram scoring, on purpose | Yes | Optional embeddings | No — substring match |
60
+ | Dependencies | **0** (stdlib `sqlite3`) | LLM client + vector DB + embedder | 0 required (v4) | Node.js runtime |
61
+
62
+ Be fair to each of them:
63
+
64
+ - **Pick Mem0** if you want automatic "talk → extracted profile" with semantic search,
65
+ mature framework integrations (LangChain, CrewAI, AutoGen) and published benchmark
66
+ numbers. That's a lot of real value. The trade-offs are documented by its own
67
+ community: an audit of 10,134 Mem0 entries found 97.8% junk, including hallucinated
68
+ user profiles that get re-extracted and amplified across sessions
69
+ ([mem0#4573](https://github.com/mem0ai/mem0/issues/4573)); the HaluMem benchmark
70
+ found extraction-stage hallucination is systemic across LLM-based memory systems
71
+ ([arXiv:2511.03506](https://arxiv.org/abs/2511.03506)). If the wording of a memory
72
+ doesn't have to be provably what a human said, that trade may be fine for you.
73
+ - **Pick memoripy** if you like the evidence-first philosophy on a **single-user**
74
+ workload and want an official MCP server out of the box.
75
+ - **Pick the official MCP memory server** if you need a minimal scratchpad for one
76
+ agent on one machine. It calls itself a reference implementation, and it is honest
77
+ about being one.
78
+ - **Pick artl-memory** if the question is *"several people talk to one agent — in a
79
+ meeting room, a family, a mediation session — and the record has to hold up later"*:
80
+ who said which exact sentence, how confident was the transcription, what did the
81
+ agent believe before the correction, and can 王强's utterances really be erased
82
+ byte-for-byte when he asks. If a smart profile matters more than a trustworthy
83
+ ledger, we are the wrong tool — and this README will tell you so again below.
84
+
85
+ ## Why multi-person
86
+
87
+ Most agent memory systems are **single-user** models: one user, one profile.
88
+ artl-memory was a multi-person model from day one:
89
+
90
+ - **Per-person isolation**: 王强's utterances never leak into 李四's recall; skills and
91
+ plugins only see their own slot through `MemView`, which holds no store handle
92
+ (same-process introspection can still dig — for hostile code, use process isolation)
93
+ - **Verbatim first**: the original sentence is stored before any judgement; rewriting
94
+ history does not exist on this path
95
+ - **Evidence chains**: every memory carries `source_event` (which utterance it came
96
+ from) and `confidence`; corrections expire the old entry and link it via
97
+ `supersedes` — **expiry is not deletion**
98
+ - **Time-based downgrade**: utterances older than N days are pruned — unreferenced
99
+ rows are deleted, referenced rows keep their snippet so the chain never dangles
100
+ - **Right to be forgotten**: `store.purge_designer("王强")` physically erases all five
101
+ tables' worth of his data (`PRAGMA secure_delete` + `VACUUM` + WAL truncation),
102
+ and returns per-table deletion counts
103
+
104
+ ## Quickstart (30 seconds)
105
+
106
+ ```python
107
+ from artl_memory import SqliteMemoryStore, MemoryService, MemView
108
+
109
+ store = SqliteMemoryStore("memory.db")
110
+ mem = MemoryService(store, designer="heima")
111
+
112
+ mem.remember("preference", "王强:预算", "三十万",
113
+ source="utterance:wq-001", confidence=0.9)
114
+ mem.recall("王强的预算") # relevance × confidence × time decay — every factor explainable
115
+
116
+ view = MemView(store, owner="skill:render", designer="heima")
117
+ view.put("风格", "奶油风") # this skill's memory; nobody else's gets in
118
+ ```
119
+
120
+ Run the multi-person example:
121
+
122
+ ```bash
123
+ python examples/two_people.py
124
+ ```
125
+
126
+ ## Zero dependencies, two backends
127
+
128
+ - **Zero hard dependencies**: standard library only (`sqlite3` is built in). A memory
129
+ layer should not drag a dependency tree through your door.
130
+ - `InMemoryStore` for tests and trials (not thread-safe); `SqliteMemoryStore` with
131
+ WAL, single connection, auto-reopen after close, corrupt-file quarantine, and a
132
+ versioned schema.
133
+ - The same conformance suite runs against both backends:
134
+ `python tests/test_conformance.py` (pytest-compatible)
135
+ - Adversarial tests cover "undocumented but must not explode":
136
+ `python tests/test_adversarial.py`
137
+ - Regression tests pin every bug ever found by external review:
138
+ `python tests/test_regressions.py`
139
+
140
+ ## Boundaries (stated plainly)
141
+
142
+ - Values are stored JSON-serialized: `tuple` comes back as `list`; serialize custom
143
+ objects yourself.
144
+ - `recall` is O(n) full-scan scoring — deliberately index-free for explainability.
145
+ The scan window is `MemoryService(recall_window=...)`, default 10000 entries; when
146
+ the window fills up you get a loud `RuntimeWarning`, never silent truncation.
147
+ For bulk imports use `remember_many(...)` (one fsync per batch, ~100x faster than
148
+ per-item commits) and make sure the window exceeds your total entry count.
149
+ - **recall scores by literal token overlap; it does not paraphrase.** Rephrasing or
150
+ mixing languages will miss: `who pays thirty万 for the living room` does not
151
+ recall the Chinese entry「三十万」. If you need semantic search, layer embeddings
152
+ on top of `recall`.
153
+ - Time decay floors at 0.4: very old high-confidence entries keep participating in
154
+ ranking, just discounted. To retire an old memory from ranking entirely, `forget`
155
+ it (expiry, not deletion).
156
+ - Empty query returns an empty list — when there is no question, it does not pretend
157
+ to understand one.
158
+ - Lock boundaries: `MemoryService.remember` is thread-safe via the service lock;
159
+ `InMemoryStore` itself is not thread-safe — use it through the service layer.
160
+ Multi-process writes to the same SQLite file are guarded by database-level unique
161
+ constraints (only one active entry per slot+key); the loser re-reads and
162
+ converges automatically.
163
+ - Isolation is at the API surface. Nothing in-process is cryptographically hidden —
164
+ if plugins are adversarial, put them behind a process boundary.
165
+
166
+ ## Explicitly not doing
167
+
168
+ - **No vector retrieval.** bigram recall is deliberate: you can always see *why* it
169
+ remembered this entry. Need embeddings? Layer them on top of `recall`.
170
+ - **No cloud sync, no multi-machine.** Your data is your file. Feature, not bug.
171
+ - **No LLM calls.** The memory layer doesn't know LLMs exist — swap brains freely,
172
+ the memory stays.
173
+
174
+ ## License
175
+
176
+ Apache-2.0 · ARTL family
@@ -0,0 +1,158 @@
1
+ # artl-memory
2
+
3
+ **The multi-person memory layer for agents.**
4
+ Verbatim utterances · per-person isolation · evidence chains · confidence scores.
5
+
6
+ > It remembers everyone in the room — separately.
7
+ > We don't promise understanding. We promise not to forget.
8
+
9
+ [中文文档](README.zh-CN.md)
10
+
11
+ artl-memory is extracted from a production assistant. It does **not** "summarize a
12
+ profile of you and pretend to know you" — it stores **who said exactly what, and when**,
13
+ each memory carrying its source utterance and a confidence score. When a fact is
14
+ corrected, the old entry is marked expired — never silently overwritten — and the
15
+ evidence chain stays in place.
16
+
17
+ > **Scope: the ledger, not the ears.** artl-memory contains no audio, no ASR, no
18
+ > voiceprint. Your upstream pipeline (diarization, voiceprint, manual labels)
19
+ > decides *who* said what and hands this library the name — it keeps the books.
20
+ > That is what makes it dependency-free: usable without ears, plug-and-play with
21
+ > them.
22
+
23
+ ```bash
24
+ pip install . # from source (PyPI package not yet published)
25
+ ```
26
+
27
+ ## Which memory layer should you pick?
28
+
29
+ Everyone arriving here has Mem0 (~60k★) as their reference point, so let's be direct
30
+ about it. These four tools answer **different questions** — this table is for picking
31
+ the right one, not for ranking them.
32
+
33
+ | | **artl-memory** | **Mem0** | **memoripy v4** | **MCP memory (official)** |
34
+ |---|---|---|---|---|
35
+ | Needs an LLM to write memory | **No** — zero model calls | Yes — extraction LLM on every `add()` | No required — deterministic extractor, bring-your-own optional | No — the agent itself decides what to store |
36
+ | What gets stored | **Verbatim utterances** + derived facts | LLM-extracted summaries | Evidence + derived memory, versioned | Entities / relations / observations |
37
+ | Can you trace a memory to its source? | Yes — `source_event` + confidence on every entry | Partial — extraction rewrites wording | Yes — citations and receipts | No |
38
+ | Multi-person isolation | **First-class** — per-person ledger, skills get scoped `MemView`s | Per-`user_id` scoping | Per-`user_id` scoping | No — one shared graph |
39
+ | Deletion | **Hard delete** — `secure_delete` + `VACUUM` + WAL truncation, verified by byte-level scan | API delete; vector-store semantics | `forget` tooling | Delete tools on the JSONL file |
40
+ | Works fully offline | Yes | No (LLM + vector store calls) | Yes | Yes |
41
+ | Semantic (vector) search | **No** — explainable bigram scoring, on purpose | Yes | Optional embeddings | No — substring match |
42
+ | Dependencies | **0** (stdlib `sqlite3`) | LLM client + vector DB + embedder | 0 required (v4) | Node.js runtime |
43
+
44
+ Be fair to each of them:
45
+
46
+ - **Pick Mem0** if you want automatic "talk → extracted profile" with semantic search,
47
+ mature framework integrations (LangChain, CrewAI, AutoGen) and published benchmark
48
+ numbers. That's a lot of real value. The trade-offs are documented by its own
49
+ community: an audit of 10,134 Mem0 entries found 97.8% junk, including hallucinated
50
+ user profiles that get re-extracted and amplified across sessions
51
+ ([mem0#4573](https://github.com/mem0ai/mem0/issues/4573)); the HaluMem benchmark
52
+ found extraction-stage hallucination is systemic across LLM-based memory systems
53
+ ([arXiv:2511.03506](https://arxiv.org/abs/2511.03506)). If the wording of a memory
54
+ doesn't have to be provably what a human said, that trade may be fine for you.
55
+ - **Pick memoripy** if you like the evidence-first philosophy on a **single-user**
56
+ workload and want an official MCP server out of the box.
57
+ - **Pick the official MCP memory server** if you need a minimal scratchpad for one
58
+ agent on one machine. It calls itself a reference implementation, and it is honest
59
+ about being one.
60
+ - **Pick artl-memory** if the question is *"several people talk to one agent — in a
61
+ meeting room, a family, a mediation session — and the record has to hold up later"*:
62
+ who said which exact sentence, how confident was the transcription, what did the
63
+ agent believe before the correction, and can 王强's utterances really be erased
64
+ byte-for-byte when he asks. If a smart profile matters more than a trustworthy
65
+ ledger, we are the wrong tool — and this README will tell you so again below.
66
+
67
+ ## Why multi-person
68
+
69
+ Most agent memory systems are **single-user** models: one user, one profile.
70
+ artl-memory was a multi-person model from day one:
71
+
72
+ - **Per-person isolation**: 王强's utterances never leak into 李四's recall; skills and
73
+ plugins only see their own slot through `MemView`, which holds no store handle
74
+ (same-process introspection can still dig — for hostile code, use process isolation)
75
+ - **Verbatim first**: the original sentence is stored before any judgement; rewriting
76
+ history does not exist on this path
77
+ - **Evidence chains**: every memory carries `source_event` (which utterance it came
78
+ from) and `confidence`; corrections expire the old entry and link it via
79
+ `supersedes` — **expiry is not deletion**
80
+ - **Time-based downgrade**: utterances older than N days are pruned — unreferenced
81
+ rows are deleted, referenced rows keep their snippet so the chain never dangles
82
+ - **Right to be forgotten**: `store.purge_designer("王强")` physically erases all five
83
+ tables' worth of his data (`PRAGMA secure_delete` + `VACUUM` + WAL truncation),
84
+ and returns per-table deletion counts
85
+
86
+ ## Quickstart (30 seconds)
87
+
88
+ ```python
89
+ from artl_memory import SqliteMemoryStore, MemoryService, MemView
90
+
91
+ store = SqliteMemoryStore("memory.db")
92
+ mem = MemoryService(store, designer="heima")
93
+
94
+ mem.remember("preference", "王强:预算", "三十万",
95
+ source="utterance:wq-001", confidence=0.9)
96
+ mem.recall("王强的预算") # relevance × confidence × time decay — every factor explainable
97
+
98
+ view = MemView(store, owner="skill:render", designer="heima")
99
+ view.put("风格", "奶油风") # this skill's memory; nobody else's gets in
100
+ ```
101
+
102
+ Run the multi-person example:
103
+
104
+ ```bash
105
+ python examples/two_people.py
106
+ ```
107
+
108
+ ## Zero dependencies, two backends
109
+
110
+ - **Zero hard dependencies**: standard library only (`sqlite3` is built in). A memory
111
+ layer should not drag a dependency tree through your door.
112
+ - `InMemoryStore` for tests and trials (not thread-safe); `SqliteMemoryStore` with
113
+ WAL, single connection, auto-reopen after close, corrupt-file quarantine, and a
114
+ versioned schema.
115
+ - The same conformance suite runs against both backends:
116
+ `python tests/test_conformance.py` (pytest-compatible)
117
+ - Adversarial tests cover "undocumented but must not explode":
118
+ `python tests/test_adversarial.py`
119
+ - Regression tests pin every bug ever found by external review:
120
+ `python tests/test_regressions.py`
121
+
122
+ ## Boundaries (stated plainly)
123
+
124
+ - Values are stored JSON-serialized: `tuple` comes back as `list`; serialize custom
125
+ objects yourself.
126
+ - `recall` is O(n) full-scan scoring — deliberately index-free for explainability.
127
+ The scan window is `MemoryService(recall_window=...)`, default 10000 entries; when
128
+ the window fills up you get a loud `RuntimeWarning`, never silent truncation.
129
+ For bulk imports use `remember_many(...)` (one fsync per batch, ~100x faster than
130
+ per-item commits) and make sure the window exceeds your total entry count.
131
+ - **recall scores by literal token overlap; it does not paraphrase.** Rephrasing or
132
+ mixing languages will miss: `who pays thirty万 for the living room` does not
133
+ recall the Chinese entry「三十万」. If you need semantic search, layer embeddings
134
+ on top of `recall`.
135
+ - Time decay floors at 0.4: very old high-confidence entries keep participating in
136
+ ranking, just discounted. To retire an old memory from ranking entirely, `forget`
137
+ it (expiry, not deletion).
138
+ - Empty query returns an empty list — when there is no question, it does not pretend
139
+ to understand one.
140
+ - Lock boundaries: `MemoryService.remember` is thread-safe via the service lock;
141
+ `InMemoryStore` itself is not thread-safe — use it through the service layer.
142
+ Multi-process writes to the same SQLite file are guarded by database-level unique
143
+ constraints (only one active entry per slot+key); the loser re-reads and
144
+ converges automatically.
145
+ - Isolation is at the API surface. Nothing in-process is cryptographically hidden —
146
+ if plugins are adversarial, put them behind a process boundary.
147
+
148
+ ## Explicitly not doing
149
+
150
+ - **No vector retrieval.** bigram recall is deliberate: you can always see *why* it
151
+ remembered this entry. Need embeddings? Layer them on top of `recall`.
152
+ - **No cloud sync, no multi-machine.** Your data is your file. Feature, not bug.
153
+ - **No LLM calls.** The memory layer doesn't know LLMs exist — swap brains freely,
154
+ the memory stays.
155
+
156
+ ## License
157
+
158
+ Apache-2.0 · ARTL family
@@ -0,0 +1,33 @@
1
+ # -*- coding: utf-8 -*-
2
+ """artl-memory — ARTL 家族的记忆层。
3
+
4
+ 逐句原话 · 按人隔离 · 证据链 · 把握度。
5
+ It remembers everyone in the room — separately.
6
+
7
+ from artl_memory import SqliteMemoryStore, InMemoryStore, MemoryService, MemView
8
+
9
+ store = SqliteMemoryStore("memory.db")
10
+ mem = MemoryService(store, designer="heima")
11
+ mem.remember("fact", "预算", "三十万", source="utterance:123", confidence=0.9)
12
+ mem.recall("客厅预算多少")
13
+ """
14
+
15
+ from .memview import MemView
16
+ from .memory_store import InMemoryStore
17
+ from .port import MemoryStore
18
+ from .service import MemoryItem, MemoryService
19
+ from .sqlite import SqliteMemoryStore
20
+ from .types import Op
21
+
22
+ __version__ = "0.2.0"
23
+
24
+ __all__ = [
25
+ "InMemoryStore",
26
+ "MemoryItem",
27
+ "MemoryService",
28
+ "MemoryStore",
29
+ "MemView",
30
+ "Op",
31
+ "SqliteMemoryStore",
32
+ "__version__",
33
+ ]