Repository navigation
First milestone release with the Phase 8 research module, a reworked home page, research-backed agent personalities, and a cleaned-up doc tree. Builds on v1.0.0 which completed the Phase 0–7 implementation plan.
Highlights
Phase 8 — Research-from-prompt module
New Step 0 entry point. Instead of uploading curated documents, describe what you want to predict and AI research agents gather source material via web search, then hand it to the existing pipeline unchanged.
Four runners:
- api (default, always available) — uses your configured LLM + ddgs for DuckDuckGo search. Works in the slim Docker image out of the box.
- claude — shells out to Claude Code CLI using `~/.claude` OAuth
- codex — shells out to OpenAI Codex CLI using `~/.codex` credentials
- kimi — shells out to Moonshot Kimi CLI
Enable with `RESEARCH_ENABLED=true` in `.env`. For CLI runners in Docker, use the `docker-compose.research.yml` overlay to mount host config dirs.
Home page mode picker
Step 1 on the home page now presents two parallel cards:
- Upload documents — PDF, Markdown, TXT, CSV (you already have source material)
- Research from prompt — describe what to predict (AI gathers material for you)
Both flows share the same prompt textarea and converge on a single Start button. The research card auto-disables when the module is turned off at the backend.
Research-backed agent personalities (carried from v1.0.0)
Agents use empirically-grounded personality frameworks instead of MBTI:
- Individuals use the Big Five / Five Factor Model — openness, conscientiousness, extraversion, agreeableness, neuroticism (each 0–100), grounded in the source material
- Institutions use an 8-archetype taxonomy (authoritative, technocratic, advocacy, commercial, community, media, academic, populist) plus 5 behavioural trait scores (formality, risk tolerance, transparency, responsiveness, ideological intensity)
- MBTI retained as a legacy display field
Both Twitter and Reddit agents now receive the full personality narrative in their system prompt (previously only Reddit got a one-sentence MBTI mention). See `docs/personality_frameworks.md` for rationale and academic citations.
Brand and visuals
- Brand rolled back to MiroFish [LemonFish] across all 8 views — MiroFish prominent with LemonFish as a gold suffix
- Yellow LemonFish logo (`MiroFish_lemonLogo.jpeg`) now reaches the Docker build (was using cached old logo)
- `index.html` title updated to "MiroFish [LemonFish]"
- Favicon switched to `lemon_fish_128.png`
- Default locale changed from Chinese to English
Docker and configuration
- `Dockerfile.slim` now installs research extras — `uv sync --frozen --extra research` pulls `ddgs` and `primp` so the api runner works without extra setup
- `.env.example` is now the authoritative environment reference — every one of the 53 env vars the backend reads is documented, organised into 11 sections covering primary LLM, per-step overrides, multi-provider pool, Zep, OASIS settings, Report Agent tuning, research module (13 vars), Flask server, and all 11 wizard-supported provider examples
Documentation
The `docs/` tree is now organised for public reference:
| Doc | For whom |
|---|---|
| ARCHITECTURE.md | Developers — full technical deep-dive |
| DETAILED_GUIDE.md | Users — pipeline walkthrough, configuration reference |
| personality_frameworks.md | Anyone — Big Five + org archetypes rationale with academic citations |
| oasis_dev.md | Contributors — OASIS engine internals, extension points |
| llm_providers.md | Operators — verified provider catalogue, updated weekly |
| llm_budget_planning.md | Operators — token consumption formulas, cost model |
| document_and_ontology_guide.md | Users — seed document quality guide |
| research_module.md | Users — Phase 8 research-from-prompt user guide |
README.md also fully rewritten for this release.
Getting started
```bash
git clone https://github.com/Lvigentini/LemonFish.git lemonfish
cd lemonfish
./setup.sh
```
The interactive setup wizard walks you through provider selection and generates a minimal `.env`. For manual configuration see `.env.example`.
Version roadmap
| Version | Contents |
|---|---|
| v0.2.0 | Phase 0 — baseline (fork, slim Docker, wizard, i18n) |
| v0.3.0 | Phase 1 — provider catalogue sync |
| v0.4.0 | Phase 2 — step-specific model routing |
| v0.5.0 | Phase 3 — token tracking + estimation |
| v0.6.0 | Phase 4 — multi-provider pool foundation |
| v0.7.0 | Phase 5 — provider capability detection |
| v0.8.0 | Phase 6 — multi-model persona assignment |
| v0.9.0 — v0.9.2 | Phase 7 — all 15 QoL items |
| v1.0.0 | Phase 7.16 — Big Five + organisational archetypes (replacing MBTI) |
| v1.0.1 (this release) | Phase 8 research module + home mode picker + doc cleanup |
Credits
LemonFish is a fork of MiroFish by BaiFu. Simulation powered by OASIS from the CAMEL-AI team.