earcon 0.1.0__tar.gz

earcon-0.1.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

earcon-0.1.0/PKG-INFO

@@ -0,0 +1,182 @@
+Metadata-Version: 2.4
+Name: earcon
+Version: 0.1.0
+Summary: A self-evolving memory proxy for OpenAI-compatible clients: any LLM app gets experience-driven learning by changing one baseURL.
+License: Apache-2.0
+Keywords: llm,agents,memory,self-improvement,jitrl,proxy,reinforcement-learning
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Provides-Extra: gateway
+Requires-Dist: fastapi>=0.100; extra == "gateway"
+Requires-Dist: uvicorn>=0.23; extra == "gateway"
+Provides-Extra: openai
+Requires-Dist: openai>=1.0; extra == "openai"
+Provides-Extra: dev
+Requires-Dist: pytest>=7.0; extra == "dev"
+Dynamic: license-file
+
+# earcon
+
+**A self-evolving memory proxy for OpenAI-compatible clients.**
+Point your app's `base_url` at earcon, keep working normally - the model
+starts learning from every session, with **zero weight updates** and
+**zero client changes beyond one URL**.
+
+- **What it is**: a local proxy that records your LLM sessions, has a
+  judge distill them into experience cards (a decision + a score), and
+  injects the relevant experience into future similar sessions.
+- **What it is not**: not fine-tuning, not another RAG memory of user
+  facts. It stores *behavioral* experience with returns - "this approach
+  worked, that one failed" - and changes decisions, not just knowledge.
+- **Lineage**: mechanisms from [JitRL: Just-In-Time Reinforcement
+  Learning](https://arxiv.org/abs/2601.18510) (ICML 2026 Spotlight):
+  non-parametric experience memory, advantage estimation, and
+  experience-driven action reweighting - realized for free-text agents.
+
+```
+your app (any OpenAI SDK / LangGraph / coding client ...)
+        │  change one base_url
+        ▼
+   earcon gateway ──① inject experience──▶ upstream LLM (frozen weights)
+        │ ② record every turn
+        ▼
+   ③ judge on session close ──▶ ④ SQLite experience cards ──▶ back to ①
+```
+
+![architecture](docs/architecture.png)
+
+## Why not just mem0 / Letta / Zep?
+
+Those store *facts* ("the user prefers concise answers") and retrieve
+them into context. earcon stores **scored behavior** and follows the
+JitRL loop:
+
+| | fact memories (mem0 et al.) | earcon |
+|---|---|---|
+| Stored unit | "user likes X" | "decision D on task T scored -3" |
+| Signal | none (storage only) | return / judge credit assignment |
+| Changes decisions? | indirectly, via recall | directly - experience is *retrieved by task similarity and weighted by outcome* |
+| Auditable | depends | one SQLite file, every card inspectable |
+| Intrusive | SDK / API changes | one `base_url` swap |
+
+## Quickstart
+
+```bash
+# legacy pip (<22) mishandles pyproject-only git installs (UNKNOWN-0.0.0):
+python3 -m pip install --upgrade pip
+
+# from GitHub (PyPI coming soon):
+python3 -m pip install "git+https://github.com/370540009gg-cmd/earcon.git"
+# extras: "earcon[gateway,openai]" via pip, or install fastapi/uvicorn yourself
+
+# terminal 1: run the gateway against any OpenAI-compatible upstream
+earcon serve --upstream https://api.openai.com/v1 \
+    --api-key $OPENAI_API_KEY --judge-model gpt-4o-mini
+
+# terminal 2: any OpenAI app, one line changed
+#   OpenAI(base_url="http://127.0.0.1:8800/v1", ...)
+python examples/quickstart_client.py
+```
+
+Work normally. Sessions close (30 min idle, or explicit
+`POST /v1/earcon/session/default/close`), the judge distills cards,
+and future similar requests arrive pre-loaded with your own track
+record:
+
+```bash
+curl http://127.0.0.1:8800/v1/earcon/stats
+# {"cards": 14, "sessions_open": 1, "inject": true}
+
+sqlite3 earcon_memory.db "SELECT task, action, G FROM cards LIMIT 5"
+```
+
+**Start in observation mode** while you assess your judge's card
+quality: `earcon serve ... --no-inject`. The judge is the quality
+ceiling of the whole system.
+
+## Does it actually work?
+
+The discrete-action path is verified in `examples/treasure_maze/` - a
+toy environment where traps sit on the intuitive routes, so avoiding
+them is *only* knowable through experience:
+
+```bash
+# control arm (frozen behavior):      ~2% win, 4.0 trap falls / episode
+python examples/treasure_maze/run.py --episodes 150 --seed 2 --no-memory
+
+# earcon learning arm:                 ~43% win (up to 100% in late windows),
+python examples/treasure_maze/run.py --episodes 150 --seed 2
+#                                      1.0 falls/episode, 10.5 steps vs 6 optimal
+```
+
+The learned advantage at the trap-adjacent cells comes out correctly
+signed - memory, not luck. See `docs/how-it-works.md` for the full
+loop, and the honest list of conditions under which it *won't* work.
+
+## How it works
+
+1. **Record** - every turn flows through the gateway transparently
+   (streaming included) into a session buffer.
+2. **Judge** - on session close, a background judge call distills 1-5
+   experience cards ("decision=...; score -3..+3").
+3. **Retrieve & inject** - the next session's first user message is
+   matched against past cards; the top matches join the system prompt.
+
+For **enumerable action spaces** (agents, tools), `earcon.core`
+implements the paper's full path: per-option logits from `logprobs`,
+advantage `A = Q - V` with small-sample shrinkage, and the closed-form
+update `z' = z + βA`. The maze example runs it end to end.
+
+Details and design corrections: `docs/how-it-works.md` · client
+integration guide: `docs/client-integration.md`.
+
+## Configuration
+
+`earcon serve --upstream URL --judge-model M [--api-key K]
+[--db FILE] [--port 8800] [--no-inject] [--session-timeout 1800]
+[--inject-top-k 5] [--judge-extra-body '{"thinking":{"type":"off"}}']`
+
+- Credentials: `EARCON_API_KEY` env var or `--api-key`. The gateway
+  rewrites client keys with its own, so clients need no real key.
+- Vendor-private judge parameters (e.g. disabling a reasoning mode for
+  the scoring call) go through `--judge-extra-body`; nothing
+  non-standard is sent by default.
+
+## Honest limitations
+
+- **The judge is the ceiling.** A shallow judge produces noisy cards;
+  start with `--no-inject` and audit them.
+- **Injection, not logits, on the main path** - most hosted gateways
+  don't expose `logprobs`, so free-text learning happens through
+  context. The full logit path needs a `logprobs`-capable upstream
+  (vLLM etc.) and lives in `earcon.core`.
+- **Cold start requires a working base.** If the model never succeeds,
+  there are no positive cards to reinforce (JitRL has the same
+  constraint). Rough sweet spot: base success in the 20-60% band.
+- **Single-user, single-process.** No multi-tenant isolation; memory is
+  one SQLite file. The session reaper is a background thread, not a
+  distributed system.
+- **Task matching is keyword-based.** Deliberately (it beat embeddings
+  in our maze experiments and needs no infra), but it's simple.
+
+## Roadmap
+
+- [ ] Hermes agent `MemoryProvider` adapter
+- [ ] Server-side logit bias via vLLM `logit_bias` (true step-level steering)
+- [ ] Card editor / curation CLI
+- [ ] Multi-namespace memory (per-project, per-tenant)
+
+## License
+
+Apache-2.0. Built on the mechanisms of [JitRL](https://arxiv.org/abs/2601.18510)
+(Barrett et al.) - if you use earcon in research, please cite the paper.

earcon-0.1.0/README.md

@@ -0,0 +1,155 @@
+# earcon
+
+**A self-evolving memory proxy for OpenAI-compatible clients.**
+Point your app's `base_url` at earcon, keep working normally - the model
+starts learning from every session, with **zero weight updates** and
+**zero client changes beyond one URL**.
+
+- **What it is**: a local proxy that records your LLM sessions, has a
+  judge distill them into experience cards (a decision + a score), and
+  injects the relevant experience into future similar sessions.
+- **What it is not**: not fine-tuning, not another RAG memory of user
+  facts. It stores *behavioral* experience with returns - "this approach
+  worked, that one failed" - and changes decisions, not just knowledge.
+- **Lineage**: mechanisms from [JitRL: Just-In-Time Reinforcement
+  Learning](https://arxiv.org/abs/2601.18510) (ICML 2026 Spotlight):
+  non-parametric experience memory, advantage estimation, and
+  experience-driven action reweighting - realized for free-text agents.
+
+```
+your app (any OpenAI SDK / LangGraph / coding client ...)
+        │  change one base_url
+        ▼
+   earcon gateway ──① inject experience──▶ upstream LLM (frozen weights)
+        │ ② record every turn
+        ▼
+   ③ judge on session close ──▶ ④ SQLite experience cards ──▶ back to ①
+```
+
+![architecture](docs/architecture.png)
+
+## Why not just mem0 / Letta / Zep?
+
+Those store *facts* ("the user prefers concise answers") and retrieve
+them into context. earcon stores **scored behavior** and follows the
+JitRL loop:
+
+| | fact memories (mem0 et al.) | earcon |
+|---|---|---|
+| Stored unit | "user likes X" | "decision D on task T scored -3" |
+| Signal | none (storage only) | return / judge credit assignment |
+| Changes decisions? | indirectly, via recall | directly - experience is *retrieved by task similarity and weighted by outcome* |
+| Auditable | depends | one SQLite file, every card inspectable |
+| Intrusive | SDK / API changes | one `base_url` swap |
+
+## Quickstart
+
+```bash
+# legacy pip (<22) mishandles pyproject-only git installs (UNKNOWN-0.0.0):
+python3 -m pip install --upgrade pip
+
+# from GitHub (PyPI coming soon):
+python3 -m pip install "git+https://github.com/370540009gg-cmd/earcon.git"
+# extras: "earcon[gateway,openai]" via pip, or install fastapi/uvicorn yourself
+
+# terminal 1: run the gateway against any OpenAI-compatible upstream
+earcon serve --upstream https://api.openai.com/v1 \
+    --api-key $OPENAI_API_KEY --judge-model gpt-4o-mini
+
+# terminal 2: any OpenAI app, one line changed
+#   OpenAI(base_url="http://127.0.0.1:8800/v1", ...)
+python examples/quickstart_client.py
+```
+
+Work normally. Sessions close (30 min idle, or explicit
+`POST /v1/earcon/session/default/close`), the judge distills cards,
+and future similar requests arrive pre-loaded with your own track
+record:
+
+```bash
+curl http://127.0.0.1:8800/v1/earcon/stats
+# {"cards": 14, "sessions_open": 1, "inject": true}
+
+sqlite3 earcon_memory.db "SELECT task, action, G FROM cards LIMIT 5"
+```
+
+**Start in observation mode** while you assess your judge's card
+quality: `earcon serve ... --no-inject`. The judge is the quality
+ceiling of the whole system.
+
+## Does it actually work?
+
+The discrete-action path is verified in `examples/treasure_maze/` - a
+toy environment where traps sit on the intuitive routes, so avoiding
+them is *only* knowable through experience:
+
+```bash
+# control arm (frozen behavior):      ~2% win, 4.0 trap falls / episode
+python examples/treasure_maze/run.py --episodes 150 --seed 2 --no-memory
+
+# earcon learning arm:                 ~43% win (up to 100% in late windows),
+python examples/treasure_maze/run.py --episodes 150 --seed 2
+#                                      1.0 falls/episode, 10.5 steps vs 6 optimal
+```
+
+The learned advantage at the trap-adjacent cells comes out correctly
+signed - memory, not luck. See `docs/how-it-works.md` for the full
+loop, and the honest list of conditions under which it *won't* work.
+
+## How it works
+
+1. **Record** - every turn flows through the gateway transparently
+   (streaming included) into a session buffer.
+2. **Judge** - on session close, a background judge call distills 1-5
+   experience cards ("decision=...; score -3..+3").
+3. **Retrieve & inject** - the next session's first user message is
+   matched against past cards; the top matches join the system prompt.
+
+For **enumerable action spaces** (agents, tools), `earcon.core`
+implements the paper's full path: per-option logits from `logprobs`,
+advantage `A = Q - V` with small-sample shrinkage, and the closed-form
+update `z' = z + βA`. The maze example runs it end to end.
+
+Details and design corrections: `docs/how-it-works.md` · client
+integration guide: `docs/client-integration.md`.
+
+## Configuration
+
+`earcon serve --upstream URL --judge-model M [--api-key K]
+[--db FILE] [--port 8800] [--no-inject] [--session-timeout 1800]
+[--inject-top-k 5] [--judge-extra-body '{"thinking":{"type":"off"}}']`
+
+- Credentials: `EARCON_API_KEY` env var or `--api-key`. The gateway
+  rewrites client keys with its own, so clients need no real key.
+- Vendor-private judge parameters (e.g. disabling a reasoning mode for
+  the scoring call) go through `--judge-extra-body`; nothing
+  non-standard is sent by default.
+
+## Honest limitations
+
+- **The judge is the ceiling.** A shallow judge produces noisy cards;
+  start with `--no-inject` and audit them.
+- **Injection, not logits, on the main path** - most hosted gateways
+  don't expose `logprobs`, so free-text learning happens through
+  context. The full logit path needs a `logprobs`-capable upstream
+  (vLLM etc.) and lives in `earcon.core`.
+- **Cold start requires a working base.** If the model never succeeds,
+  there are no positive cards to reinforce (JitRL has the same
+  constraint). Rough sweet spot: base success in the 20-60% band.
+- **Single-user, single-process.** No multi-tenant isolation; memory is
+  one SQLite file. The session reaper is a background thread, not a
+  distributed system.
+- **Task matching is keyword-based.** Deliberately (it beat embeddings
+  in our maze experiments and needs no infra), but it's simple.
+
+## Roadmap
+
+- [ ] Hermes agent `MemoryProvider` adapter
+- [ ] Server-side logit bias via vLLM `logit_bias` (true step-level steering)
+- [ ] Card editor / curation CLI
+- [ ] Multi-namespace memory (per-project, per-tenant)
+
+## License
+
+Apache-2.0. Built on the mechanisms of [JitRL](https://arxiv.org/abs/2601.18510)
+(Barrett et al.) - if you use earcon in research, please cite the paper.

earcon-0.1.0/earcon/__init__.py

@@ -0,0 +1,10 @@
+"""Earcon: a self-evolving memory proxy for OpenAI-compatible clients.
+
+Any LLM application gets experience-driven learning by changing one
+baseURL. Model weights stay frozen; experience is accumulated,
+adjudicated, and injected as context. See docs/how-it-works.md.
+
+Based on the mechanisms of JitRL (arXiv:2601.18510).
+"""
+
+__version__ = "0.1.0"

earcon-0.1.0/earcon/cli.py

@@ -0,0 +1,72 @@
+# -*- coding: utf-8 -*-
+"""CLI: `earcon serve` starts the learning proxy.
+
+Examples:
+    earcon serve --upstream https://api.openai.com/v1 --api-key $KEY \
+        --judge-model gpt-4o-mini --port 8800
+
+    # pure observation (no injection) while you assess judge quality:
+    earcon serve --upstream ... --no-inject
+"""
+
+import argparse
+import json
+import os
+
+
+def build_parser():
+    p = argparse.ArgumentParser(prog="earcon", description=__doc__)
+    sub = p.add_subparsers(dest="cmd")
+    serve = sub.add_parser("serve", help="run the learning proxy gateway")
+    serve.add_argument("--upstream", required=True,
+                       help="OpenAI-compatible upstream base URL, e.g. "
+                            "https://api.openai.com/v1 "
+                            "(env: EARCON_UPSTREAM)")
+    serve.add_argument("--api-key", default=os.environ.get("EARCON_API_KEY", ""),
+                       help="key used for upstream calls; clients' own keys "
+                            "are ignored (env: EARCON_API_KEY)")
+    serve.add_argument("--port", type=int, default=8800)
+    serve.add_argument("--db", default="earcon_memory.db")
+    serve.add_argument("--judge-model", required=True,
+                       help="model used for credit assignment")
+    serve.add_argument("--no-inject", action="store_true",
+                       help="record-only mode: judge sessions, never inject")
+    serve.add_argument("--judge-extra-body", default=None,
+                       help="JSON object merged into judge requests, for "
+                            "vendor-private params (e.g. disabling a "
+                            "reasoning mode). Never sent to the main path.")
+    serve.add_argument("--session-timeout", type=float, default=1800,
+                       help="seconds of inactivity before a session is "
+                            "closed and judged (default 1800)")
+    serve.add_argument("--inject-top-k", type=int, default=5)
+    return p
+
+
+def main(argv=None):
+    p = build_parser()
+    args = p.parse_args(argv)
+    if args.cmd != "serve":
+        p.print_help()
+        return 2
+
+    extra = json.loads(args.judge_extra_body) if args.judge_extra_body else None
+
+    # local imports so the core library never requires fastapi
+    from earcon.gateway import Gateway, create_app
+    gw = Gateway(upstream=args.upstream, api_key=args.api_key,
+                 db_path=args.db, judge_model=args.judge_model,
+                 inject=not args.no_inject, extra_body=extra,
+                 config={"session_timeout": args.session_timeout,
+                         "inject_top_k": args.inject_top_k})
+    app = create_app(gw)
+
+    import uvicorn
+    print("earcon gateway: http://127.0.0.1:%d  ->  %s" % (args.port, args.upstream))
+    print("memory: %s | mode: %s" % (args.db,
+                                     "record+inject" if gw.inject else "record-only"))
+    uvicorn.run(app, host="127.0.0.1", port=args.port, log_level="warning")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

earcon-0.1.0/earcon/core/__init__.py

@@ -0,0 +1,5 @@
+"""Core library: memory, advantage estimation, policies, judges, backends.
+
+Pure logic, zero third-party dependencies (the OpenAI-compatible backend
+lazily imports `openai` only when used).
+"""