Refresh README for v0.7.1 direction

Author: jeong87

CHANGELOG.md

@@ -6,6 +6,32 @@ Note:
 - Repository history does not contain a public `0.6.0` version bump.
 - The preview-first and local-inbox transition work from that period was folded into `0.7.0`.
 
+## [0.7.1] - 2026-04-03
+
+### Added
+- Added the first complete local agent interface for `2-2`:
+  - local-only HTTP JSON endpoint: `POST /api/agent/search`
+  - CLI JSON mode: `python app/paper_digest_app.py --agent-search ...`
+  - separate `AGENT_API_TOKEN` broker token so agents do not need raw provider keys
+- Added OPENAI-compatible local/self-hosted backend support for agent-oriented use:
+  - `ENABLE_OPENAI_COMPAT_FALLBACK`
+  - `OPENAI_COMPAT_API_BASE`
+  - `OPENAI_COMPAT_MODEL`
+  - `OPENAI_COMPAT_API_KEY`
+- Added dedicated agent manuals:
+  - `docs/manuals/MANUAL_AGENT_EN.md`
+  - `docs/manuals/MANUAL_AGENT_KR.md`
+
+### Changed
+- Reorganized the English and Korean README front sections around the current product direction:
+  - `2-1` human on-demand search
+  - `2-2` agent-oriented search
+- Moved older delivery/automation paths such as email and GitHub Actions lower in the README as optional workflows.
+- Settings UI now exposes agent-mode controls and OPENAI-compatible backend settings.
+
+### Docs
+- Updated README, Korean README, and manual cross-links to reflect the search-first + agent-tool product framing.
+
 ## [0.7.0] - 2026-04-02
 
 ### Added

README.md

@@ -1,13 +1,13 @@
-# Paper Morning
+# Paper Morning
 
 [![Paper Morning Logo](assets/papermorning2.png)](https://raw.githack.com/jeong87/paper-morning/main/docs/preview/index.html)
 
 **[EN](README.md) | [KR](docs/manuals/README_KR.md)**
 
-Paper Morning is a research-context paper search tool for medical/health AI researchers.
-It turns a project description into search queries, retrieves papers, ranks them by practical relevance, and explains why they matter. Local inbox and scheduled digest delivery remain available as optional follow-up modes.
+Paper Morning is a research-context paper search engine for medical and health AI work.
+It turns a project description into search queries, retrieves papers, reranks them by practical relevance, and explains why they matter.
 
-- Latest version: **[v0.7.0](VERSION)**
+- Latest version: **[v0.7.1](VERSION)**
 - License: `GNU AGPLv3` ([LICENSE](LICENSE))
 - Privacy policy: [PRIVACY.md](PRIVACY.md)
 
@@ -29,100 +29,131 @@ Notes:
 - No actual email is sent from this page
 - Best for first-impression validation before local install
 
-## What It Does
-1. Reads your project context and saved search queries.
-2. Collects papers from arXiv, PubMed, Semantic Scholar, and optional Google Scholar (SerpAPI).
-3. Scores each paper (1-10) with LLM relevance ranking.
-4. Lets you inspect results on demand, then optionally save or deliver them later.
-
-## Key Features
-- Personalized LLM relevance ranking using your active projects.
-- Search intents for recent updates, best-fit papers, and broader discovery.
-- Local agent interface with brokered credentials and JSON output.
-- Per-project mail cadence (`daily` / `every_3_days` / `weekly`) in Topic Editor.
-- Local inbox mode with browser popup scheduling on a running PC.
-- Configurable cadence: `daily`, `every_3_days`, `weekly`.
-- Duplicate suppression with history tracking (`sent_ids.json`).
-- PubMed 429 retry/backoff handling.
-- Gemini with automatic fallback (`3.1-pro` -> `3.1-flash` -> `3.0-pro` -> `3.0-flash` -> `2.5-pro` -> `2.5-flash`) and optional Cerebras backup.
-- Output language control via `.env`:
-  - `OUTPUT_LANGUAGE=en|ko|ja|es|...`
-
-## Quick Start
-- Beginner (English): [docs/manuals/MANUAL_FIRSTTIME_EN.md](docs/manuals/MANUAL_FIRSTTIME_EN.md)
-- Full operations (English): [docs/manuals/MANUAL_EN.md](docs/manuals/MANUAL_EN.md)
-- Agent/tool integration (English): [docs/manuals/MANUAL_AGENT_EN.md](docs/manuals/MANUAL_AGENT_EN.md)
-- Beginner (Korean): [docs/manuals/MANUAL_FIRSTTIME_KR.md](docs/manuals/MANUAL_FIRSTTIME_KR.md)
-- Full operations (Korean): [docs/manuals/MANUAL_KR.md](docs/manuals/MANUAL_KR.md)
-- Agent/tool integration (Korean): [docs/manuals/MANUAL_AGENT_KR.md](docs/manuals/MANUAL_AGENT_KR.md)
+## Product Direction
+Paper Morning is no longer centered on "send me papers every morning."
+
+The core product is now:
+- a human-facing on-demand paper search tool
+- a local agent-facing paper search tool with structured JSON output
+
+The old digest and email workflows still exist, but they are now optional follow-up features.
+
+## 2-1 Human Search
+Use Paper Morning when you want to check for relevant papers on demand.
+
+What it supports:
+- `What's New`: recent updates first
+- `Best Match`: strongest-fit papers in the selected horizon
+- `Discovery`: adjacent but transferable work
+- Time horizons from `7d` to `5y`
+- Local inbox storage and browser-based result viewing
+
+Best fit for:
+- researchers exploring a new direction
+- people who want context-fit papers, not just the newest papers
+- users who do not want to set up email first
 
-## Recommended First Path (Search-First, Local)
-Generate your first personalized search result before email or automation setup.
+## 2-2 Agent Search
+Paper Morning can also act as a local paper-search tool for research agents.
 
-1. Install dependencies:
+What it supports:
+- local HTTP JSON endpoint: `POST /api/agent/search`
+- CLI JSON mode: `python app/paper_digest_app.py --agent-search ...`
+- brokered local auth via `AGENT_API_TOKEN`
+- Gemini or a local/self-hosted OPENAI-compatible backend
+
+Security model:
+- your agent gets only `AGENT_API_TOKEN`
+- Paper Morning keeps `GEMINI_API_KEY` or local backend credentials in `.env` or OS keyring
+- the agent never needs to see raw provider keys
+
+Best fit for:
+- literature review agents
+- planning / scouting agents
+- local tool pipelines that need structured paper search output
+
+## How It Works
+Shared engine for both human and agent paths:
+
+1. Read research context
+2. Generate source-specific search queries
+3. Retrieve candidates from arXiv, PubMed, Semantic Scholar, and optional Google Scholar
+4. Apply intent-aware horizon filtering
+5. Build a shortlist
+6. Run listwise LLM reranking across the shortlist
+7. Return either:
+   - HTML result view for humans
+   - structured JSON for agents
+
+## Start Here
+- Live preview first: [Open Live Web Preview](https://raw.githack.com/jeong87/paper-morning/main/docs/preview/index.html)
+- Human local usage: [docs/manuals/MANUAL_FIRSTTIME_EN.md](docs/manuals/MANUAL_FIRSTTIME_EN.md)
+- Agent/tool usage: [docs/manuals/MANUAL_AGENT_EN.md](docs/manuals/MANUAL_AGENT_EN.md)
+- Korean README: [docs/manuals/README_KR.md](docs/manuals/README_KR.md)
+
+## Optional Workflows
+These are still supported, but they are no longer the main product story.
+
+### Local install and local UI
+If you want to actually use Paper Morning on your machine:
 
 ```bash
 pip install -r deps/requirements.txt
+python app/local_ui_launcher.py
 ```
 
-2. Run local launcher:
+Main local UI:
 
-```bash
-python app/local_ui_launcher.py
+```text
+http://127.0.0.1:5050
 ```
 
-3. The browser opens automatically. On first run, Setup Wizard opens automatically.
-4. Fill project description + Gemini key, then click `Save and Search Now`.
+### Local inbox and morning popup
+You can keep the app running and let it open the default search result automatically at your scheduled time.
+
+Relevant settings:
+- `DELIVERY_MODE=local_inbox`
+- `AUTO_OPEN_DIGEST_WINDOW=true`
+- `SEARCH_INTENT_DEFAULT`
+- `SEARCH_TIME_HORIZON_DEFAULT`
+
+### Email delivery
+Email is optional.
 
-This verifies product value first without Gmail or GitHub Actions setup. After setup, the default path is:
-- Click one button to search and open the latest result in a browser tab
-- Or keep the local UI running and let the scheduled morning popup open automatically
+Priority:
+1. Local Inbox
+2. Gmail OAuth
+3. Gmail App Password
 
-## GitHub Actions Mode (Advanced Automation)
-Use this only after local search quality is confirmed and you really want automation.
+Docs:
+- Gmail app password: https://myaccount.google.com/apppasswords
+- Full ops manual: [docs/manuals/MANUAL_EN.md](docs/manuals/MANUAL_EN.md)
+
+### GitHub Actions automation
+Use this only if you really want unattended automation without relying on a running local PC.
 
 Required workflow files:
 - `.github/workflows/paper-morning-digest.yml`
 - `.github/workflows/paper-morning-bootstrap-topics.yml`
 
-Required secrets:
-- `PM_ENV_FILE` (full `.env` content)
+Required secret:
+- `PM_ENV_FILE`
 
-Optional secret:
-- `PM_TOPICS_JSON` (full `user_topics.json` content)
-- `PM_PROJECTS_JSON` (project list only, for bootstrap query generation)
+Optional secrets:
+- `PM_TOPICS_JSON`
+- `PM_PROJECTS_JSON`
 
 Tracked non-secret config:
-- `config/projects.yaml` (project descriptions used for bootstrap/default onboarding)
+- `config/projects.yaml`
 
 ## Important Settings
-- `ONBOARDING_MODE`: `preview` (default) or `daily`.
-- `SEARCH_INTENT_DEFAULT`: default local search mode (`best_match`, `whats_new`, `discovery`).
-- `SEARCH_TIME_HORIZON_DEFAULT`: default local search horizon (`7d`, `30d`, `180d`, `1y`, `3y`, `5y`).
-- `SEND_FREQUENCY` / `SEND_ANCHOR_DATE`: cadence policy.
-- `LOOKBACK_HOURS`: legacy fallback window for digest-style collection.
-- `LLM_MAX_CANDIDATES`: shortlist cap for listwise LLM reranking.
-- `OUTPUT_LANGUAGE`: summary language for LLM-generated reason/core/usefulness text.
-- `ENABLE_GOOGLE_SCHOLAR` + `GOOGLE_SCHOLAR_API_KEY`: optional SerpAPI source.
-
-## Local Web Console
-Main path for onboarding and preview-first setup:
-
-```text
-http://127.0.0.1:5050
-```
-
-## Agent Mode
-Paper Morning can also be used as a local paper-search tool for research agents.
-
-Security model:
-- your agent gets `AGENT_API_TOKEN`
-- Paper Morning keeps `GEMINI_API_KEY` or local backend credentials in `.env` / keyring
-- the agent never needs to see raw provider keys
-
-Interfaces:
-- Local HTTP JSON endpoint: `POST /api/agent/search`
-- CLI JSON mode: `python app/paper_digest_app.py --agent-search ...`
+- `SEARCH_INTENT_DEFAULT`: default local search mode
+- `SEARCH_TIME_HORIZON_DEFAULT`: default local search horizon
+- `LLM_MAX_CANDIDATES`: shortlist cap for listwise reranking
+- `OUTPUT_LANGUAGE`: language for LLM-generated explanation fields
+- `AGENT_API_TOKEN`: local broker token for agent mode
+- `ENABLE_OPENAI_COMPAT_FALLBACK`: enable LM Studio / vLLM / other OpenAI-style local backend
+- `ENABLE_GOOGLE_SCHOLAR` + `GOOGLE_SCHOLAR_API_KEY`: optional SerpAPI source
 
 ## Build Distribution Files
 ### Windows
@@ -142,45 +173,36 @@ This repo includes:
 - pages workflow: `.github/workflows/deploy-demo-pages.yml`
 
 To publish demo on your fork:
-1. Enable GitHub Pages source as **GitHub Actions**.
-2. Run `deploy-demo-pages` workflow (or push to `main`).
-
-Repo settings TODO (outside code):
-- Fill GitHub About fields (`description`, `website`, `topics`).
-- Publish the first tagged GitHub Release.
+1. Enable GitHub Pages source as **GitHub Actions**
+2. Run `deploy-demo-pages` workflow or push to `main`
 
 ## Template-First Repository Setup
-For global onboarding, prefer:
+For onboarding, prefer:
 - `Use this template` -> create your own repository instance
 
-Fallback (advanced):
-- fork workflow if you explicitly need upstream fork linkage.
+Fallback:
+- fork only if you explicitly need upstream fork linkage
 
 ## Actions Cost Note (Private Repos)
-- GitHub Free private repos include limited Actions minutes.
-- If your workflow runs frequently (for example every 15 minutes), usage can exceed free minutes.
-- Local-first preview and local scheduling are recommended for cost-safe onboarding.
+- GitHub Free private repos include limited Actions minutes
+- frequent automation can exceed the free tier quickly
+- local-first search and local scheduling remain the cheaper default path
 
 ## Troubleshooting (Quick)
-- `Search query is empty`: generate and save topics/queries in Topic Editor.
-- `PubMed 429`: retries are automatic, but adding `NCBI_API_KEY` is recommended.
-- `Gemini model 404 / quota`: the live preview now retries multiple Gemini fallbacks automatically before surfacing an error.
-- No email received: check sender/recipient addresses, spam folder, and auth config.
-
-## Delivery Priority
-1. Local Inbox (default, no email credentials required)
-2. Gmail OAuth (optional advanced mode)
-3. Gmail App Password (optional fallback)
-
-Gmail app password docs:
-- https://myaccount.google.com/apppasswords
+- `Search query is empty`: generate and save topics/queries in Topic Editor
+- `PubMed 429`: retries are automatic, but adding `NCBI_API_KEY` is recommended
+- `Gemini model 404 / quota`: the preview and runtime try multiple Gemini fallbacks before failing
+- Agent `403 Forbidden`: check `AGENT_API_TOKEN` and confirm the request is coming from the local machine
+- No email received: check sender/recipient addresses, spam folder, and auth config
 
 ## Documentation
 - Beginner (English): [docs/manuals/MANUAL_FIRSTTIME_EN.md](docs/manuals/MANUAL_FIRSTTIME_EN.md)
 - Full operations (English): [docs/manuals/MANUAL_EN.md](docs/manuals/MANUAL_EN.md)
+- Agent/tool integration (English): [docs/manuals/MANUAL_AGENT_EN.md](docs/manuals/MANUAL_AGENT_EN.md)
 - Beginner (Korean): [docs/manuals/MANUAL_FIRSTTIME_KR.md](docs/manuals/MANUAL_FIRSTTIME_KR.md)
 - Full operations (Korean): [docs/manuals/MANUAL_KR.md](docs/manuals/MANUAL_KR.md)
-- Korean README (legacy): [docs/manuals/README_KR.md](docs/manuals/README_KR.md)
+- Agent/tool integration (Korean): [docs/manuals/MANUAL_AGENT_KR.md](docs/manuals/MANUAL_AGENT_KR.md)
+- Korean README: [docs/manuals/README_KR.md](docs/manuals/README_KR.md)
 
 ## Contact
 - nineclas@gmail.com

VERSION

@@ -1 +1 @@
-0.7.0
+0.7.1