neocore-memory 0.3.2__tar.gz → 0.3.4__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.
Files changed (88) hide show
  1. neocore_memory-0.3.4/.agents/plugins/marketplace.json +20 -0
  2. neocore_memory-0.3.4/.claude-plugin/marketplace.json +14 -0
  3. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/PKG-INFO +31 -2
  4. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/README.md +30 -1
  5. neocore_memory-0.3.4/STATUS.md +27 -0
  6. neocore_memory-0.3.4/docs/research/2026-11-19-foreground-recall.json +977 -0
  7. neocore_memory-0.3.4/docs/research/2026-11-19-foreground-recall.md +79 -0
  8. neocore_memory-0.3.4/docs/revision-authority.md +40 -0
  9. neocore_memory-0.3.4/integrations/claude-code/.claude-plugin/plugin.json +18 -0
  10. neocore_memory-0.3.4/integrations/claude-code/.mcp.json +10 -0
  11. neocore_memory-0.3.4/integrations/claude-code/hooks/hooks.json +27 -0
  12. neocore_memory-0.3.4/integrations/codex/.codex-plugin/plugin.json +36 -0
  13. neocore_memory-0.3.4/integrations/codex/.mcp.json +12 -0
  14. neocore_memory-0.3.4/integrations/codex/hooks.json +26 -0
  15. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/integrations/hermes/README.md +2 -2
  16. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/pyproject.toml +1 -1
  17. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/__init__.py +1 -1
  18. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/assistant/memory.py +66 -21
  19. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/assistant/profile.py +113 -27
  20. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/assistant/render.py +74 -64
  21. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/cli.py +17 -0
  22. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/neosift_server.py +31 -10
  23. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/recall.py +370 -110
  24. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/shared/tools.py +21 -3
  25. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/store.py +92 -20
  26. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/tests/test_assistant.py +7 -4
  27. neocore_memory-0.3.4/tests/test_erased_revision.py +204 -0
  28. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/tests/test_neosift.py +53 -0
  29. neocore_memory-0.3.4/tests/test_plugins.py +69 -0
  30. neocore_memory-0.3.4/tests/test_profile_authority.py +151 -0
  31. neocore_memory-0.3.4/tests/test_prompt_authority.py +65 -0
  32. neocore_memory-0.3.4/tests/test_recall_index_changes.py +75 -0
  33. neocore_memory-0.3.4/tests/test_revision_authority.py +137 -0
  34. neocore_memory-0.3.4/tests/test_search_window.py +62 -0
  35. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/tests/test_shared.py +23 -2
  36. neocore_memory-0.3.4/tests/test_span_neighborhood.py +68 -0
  37. neocore_memory-0.3.4/tests/test_vector_replacement.py +100 -0
  38. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/.github/workflows/publish.yml +0 -0
  39. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/.github/workflows/test.yml +0 -0
  40. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/.gitignore +0 -0
  41. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/LICENSE +0 -0
  42. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/NOTICE +0 -0
  43. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/conftest.py +0 -0
  44. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/docs/governance.md +0 -0
  45. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/docs/how-it-works.md +0 -0
  46. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/docs/shared-memory.md +0 -0
  47. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/evals/README.md +0 -0
  48. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/evals/common.py +0 -0
  49. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/evals/governance.py +0 -0
  50. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/evals/locomo.py +0 -0
  51. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/evals/longmemeval.py +0 -0
  52. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/evals/results/governance.json +0 -0
  53. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/evals/results/locomo.json +0 -0
  54. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/evals/results/longmemeval-s.json +0 -0
  55. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/integrations/hermes/__init__.py +0 -0
  56. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/integrations/hermes/plugin.yaml +0 -0
  57. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/integrations/hermes/pyproject.toml +0 -0
  58. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/integrations/hermes/tests/test_provider.py +0 -0
  59. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/integrations/hermes/tests/test_shared_memory.py +0 -0
  60. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/scripts/install/install.ps1 +0 -0
  61. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/scripts/install/install.sh +0 -0
  62. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/__main__.py +0 -0
  63. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/assistant/__init__.py +0 -0
  64. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/assistant/filter.py +0 -0
  65. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/assistant/lessons.py +0 -0
  66. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/assistant/neosift.py +0 -0
  67. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/assistant/outcomes.py +0 -0
  68. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/assistant/rules.py +0 -0
  69. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/assistant/share.py +0 -0
  70. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/assistant/sleep.py +0 -0
  71. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/consolidation.py +0 -0
  72. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/embedder.py +0 -0
  73. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/llm.py +0 -0
  74. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/shared/__init__.py +0 -0
  75. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/shared/admin.py +0 -0
  76. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/shared/client.py +0 -0
  77. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/shared/keys.py +0 -0
  78. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/shared/manage.py +0 -0
  79. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/shared/server.py +0 -0
  80. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/src/neocore/temporal.py +0 -0
  81. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/tests/test_child_env.py +0 -0
  82. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/tests/test_consolidation.py +0 -0
  83. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/tests/test_hermes_backend.py +0 -0
  84. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/tests/test_recall.py +0 -0
  85. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/tests/test_rules.py +0 -0
  86. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/tests/test_setup.py +0 -0
  87. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/tests/test_share.py +0 -0
  88. {neocore_memory-0.3.2 → neocore_memory-0.3.4}/tests/test_stuck_offsets.py +0 -0
@@ -0,0 +1,20 @@
1
+ {
2
+ "name": "neocore",
3
+ "interface": {
4
+ "displayName": "NeoCore"
5
+ },
6
+ "plugins": [
7
+ {
8
+ "name": "neocore",
9
+ "source": {
10
+ "source": "local",
11
+ "path": "./integrations/codex"
12
+ },
13
+ "policy": {
14
+ "installation": "AVAILABLE",
15
+ "authentication": "ON_INSTALL"
16
+ },
17
+ "category": "Developer Tools"
18
+ }
19
+ ]
20
+ }
@@ -0,0 +1,14 @@
1
+ {
2
+ "name": "neocore",
3
+ "owner": {
4
+ "name": "VincentPorquet777"
5
+ },
6
+ "description": "NeoCore: one memory for Claude Code, Codex and Hermes.",
7
+ "plugins": [
8
+ {
9
+ "name": "neocore",
10
+ "source": "./integrations/claude-code",
11
+ "description": "One memory for every AI you use: Claude Code, Codex and Hermes share it. Before each prompt, NeoCore adds the dated excerpts and facts that prompt needs, with zero model calls at recall."
12
+ }
13
+ ]
14
+ }
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.5
2
2
  Name: neocore-memory
3
- Version: 0.3.2
3
+ Version: 0.3.4
4
4
  Summary: Governed long-term memory for AI assistants: recalls only what the moment needs. Shared memory for Claude Code and Codex.
5
5
  Project-URL: Homepage, https://github.com/VincentPorquet777/neocore
6
6
  Author: Vincent Porquet
@@ -96,12 +96,41 @@ Codex will ask you to trust the new hooks the first time they run. `neocore unin
96
96
  the hooks and keeps your memory. `neocore doctor` checks the install. Run
97
97
  `neocore setup --dry-run` to see the plan first.
98
98
 
99
+ ### Or as a plugin
100
+
101
+ The same hooks and memory tools also come as a Claude Code plugin and a Codex plugin. Install
102
+ the package first (the plugins run the `neocore` command), then set up the memory folder
103
+ without touching your settings files:
104
+
105
+ ```bash
106
+ pipx install "neocore-memory[embed]"
107
+ neocore setup --no-claude --no-codex
108
+ ```
109
+
110
+ Claude Code:
111
+
112
+ ```
113
+ /plugin marketplace add VincentPorquet777/neocore
114
+ /plugin install neocore@neocore
115
+ ```
116
+
117
+ Codex:
118
+
119
+ ```bash
120
+ codex plugin marketplace add VincentPorquet777/neocore
121
+ codex plugin add neocore@neocore
122
+ ```
123
+
124
+ Pick one way per tool. If `neocore setup` already added its hooks, the plugin's hooks stay quiet,
125
+ so each prompt still gets its memory once.
126
+
99
127
  After setup:
100
128
 
101
129
  - **Every prompt** gets the memory that matters, as hook context, in well under a second. A warm
102
130
  local daemon serves it. A chat's first prompt also gets your briefing.
103
131
  - **Memory tools** for the AI to look further on purpose: `neocore_search`, `neocore_open`,
104
- `neocore_timeline`, `neocore_facts`, `neocore_profile` and `neocore_remember`.
132
+ `neocore_timeline`, `neocore_facts`, `neocore_profile` and `neocore_remember`. Use `kind=turns` on
133
+ `neocore_search` for the original wording of a decision. The default lists facts first.
105
134
  - **Every finished reply** starts a background sync. It reads new turns from Claude Code and
106
135
  Codex transcripts (prompts and final replies only; tool output is skipped and obvious secrets
107
136
  are redacted) and indexes them.
@@ -72,12 +72,41 @@ Codex will ask you to trust the new hooks the first time they run. `neocore unin
72
72
  the hooks and keeps your memory. `neocore doctor` checks the install. Run
73
73
  `neocore setup --dry-run` to see the plan first.
74
74
 
75
+ ### Or as a plugin
76
+
77
+ The same hooks and memory tools also come as a Claude Code plugin and a Codex plugin. Install
78
+ the package first (the plugins run the `neocore` command), then set up the memory folder
79
+ without touching your settings files:
80
+
81
+ ```bash
82
+ pipx install "neocore-memory[embed]"
83
+ neocore setup --no-claude --no-codex
84
+ ```
85
+
86
+ Claude Code:
87
+
88
+ ```
89
+ /plugin marketplace add VincentPorquet777/neocore
90
+ /plugin install neocore@neocore
91
+ ```
92
+
93
+ Codex:
94
+
95
+ ```bash
96
+ codex plugin marketplace add VincentPorquet777/neocore
97
+ codex plugin add neocore@neocore
98
+ ```
99
+
100
+ Pick one way per tool. If `neocore setup` already added its hooks, the plugin's hooks stay quiet,
101
+ so each prompt still gets its memory once.
102
+
75
103
  After setup:
76
104
 
77
105
  - **Every prompt** gets the memory that matters, as hook context, in well under a second. A warm
78
106
  local daemon serves it. A chat's first prompt also gets your briefing.
79
107
  - **Memory tools** for the AI to look further on purpose: `neocore_search`, `neocore_open`,
80
- `neocore_timeline`, `neocore_facts`, `neocore_profile` and `neocore_remember`.
108
+ `neocore_timeline`, `neocore_facts`, `neocore_profile` and `neocore_remember`. Use `kind=turns` on
109
+ `neocore_search` for the original wording of a decision. The default lists facts first.
81
110
  - **Every finished reply** starts a background sync. It reads new turns from Claude Code and
82
111
  Codex transcripts (prompts and final replies only; tool output is skipped and obvious secrets
83
112
  are redacted) and indexes them.
@@ -0,0 +1,27 @@
1
+ # Foreground recall R&D handoff
2
+
3
+ Updated 2026-11-19 03:09 UTC. Branch `codex/recall-efficiency-20261119`. Inspection state **SAFE_TO_INSPECT**. Product checkpoint `3e030c48ce333c9f346133342fe1624b90524d16`. Phase **two-hour R&D handoff, candidate for a Linux canary**.
4
+
5
+ | Owner question | Measured answer |
6
+ |---|---|
7
+ | Did useful memory arrive faster? | With the deployed policy file and local r10 model, median 2.525 to 1.660 seconds, P95 3.808 to 2.389 seconds on 64 historical queries. |
8
+ | Was a comparable retrieval preserved? | 709 of 709 complete Linux pre-scoring packets and receipts identical. Candidate faster on 706. |
9
+ | Did it fix a user's decision? | Two of two blinded synthetic decisions could cite the authorized replacement, versus zero before. One prohibited control remained unanswered. Real-user downstream benefit is unmeasured. |
10
+ | Did authority survive? | Zero leaks on 1,540 governed questions. The control leaked on all 1,540. Final outputs and receipts matched the baseline in both arms. |
11
+ | What remains weak? | Default search surfaced the original instruction directly on zero of twelve naturally occurring correction queries. Explicit turn search reached five. A proposed mixed default lost two fact-only cases and was rejected. |
12
+ | What is live? | Read-only inspection found b143.dev1 and both server and scorer active. No release, restart or new model training took place. |
13
+
14
+ ## Verification and failures
15
+
16
+ All 205 public tests passed. On the private branch all 123 focused tests passed on Windows and Linux. The installed-policy wheel passed 58 Linux tests using the VPS's actual SQLite library (3.45.1, hash recorded), including source withdrawal, current revision, isolation and cold vector replacement. The corrected indexing order recovered all five annotated support cases lost in the intermediate candidate. The project-origin diagnostic identified 346 mismatching old display labels across 9,884 historical Evidence rows.
17
+
18
+ The initial Linux broad run hit its bound. Its continuation passed 266 of 267 exercised nodes. An inherited Windows path-format expectation under simulated Linux remained. Windows broad suite has 2,066 passes, 3 failures and 12 skips recorded through its bounded run. It did not produce a complete final-source result. The slow ingestion case also failed on the reference. A stale admin test call was repaired at `cfc538c2b39473af6814ff7517f0d9c903d14a6d` in the private repository. The dependency installation test passed after its declared Pillow version was cached for offline use. Windows strict typing retains 21 inherited errors against 24 at the reference. No new added-line Ruff findings.
19
+
20
+ ## Next action and safety
21
+
22
+ 1. Canary the exact wheel on a separate, production-like Linux instance with the installed policy, actual scorer and a fresh Hermes session. Measure transport and end-to-end latency, correction and revocation before promotion.
23
+ 2. Investigate whether the model chooses original-turn search for prior instructions while preserving settled-fact performance. Keep threshold and scoring policy separate from the speed-only adoption decision.
24
+
25
+ The research fixtures used local compute, private read-only snapshots and synthetic control cases. Generated copies of the historical databases were removed. No metered model call, Fable call, RunPod connection, live write or model deployment was needed. Worktree changes for this handoff are restricted to the reports and this status, committed together. No deployment worker or continuing R&D automation was introduced.
26
+
27
+ [Read the full report](docs/research/2026-11-19-foreground-recall.md) · [Aggregate results and wheel manifests](docs/research/2026-11-19-foreground-recall.json)