CogneeMind: Agentic Algorithm Discovery for Hard Optimization Problems

.gitignore

@@ -0,0 +1,84 @@
+# ─── secrets ──────────────────────────────────────────────────────────
+# NEVER commit any env or credential file. Track .env.example instead.
+.env
+.env.*
+!.env.example
+*.pem
+*.key
+*.crt
+secrets/
+credentials.json
+service-account*.json
+
+# ─── python ───────────────────────────────────────────────────────────
+.venv/
+venv/
+env/
+__pycache__/
+*.py[cod]
+*$py.class
+*.egg
+*.egg-info/
+.eggs/
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+.coverage
+.coverage.*
+htmlcov/
+build/
+develop-eggs/
+.installed.cfg
+.tox/
+.nox/
+pip-log.txt
+pip-delete-this-directory.txt
+
+# ─── node / vite / frontend build ─────────────────────────────────────
+node_modules/
+dist/
+dist-ssr/
+.vite/
+.cache/
+.parcel-cache/
+.turbo/
+*.tsbuildinfo
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+pnpm-debug.log*
+.pnpm-store/
+
+# ─── runtime / generated data ─────────────────────────────────────────
+# Wiki + evidence are regenerated at runtime — keep them out of git.
+backend/data/wiki*.json
+backend/data/evidence.json
+backend/data/wiki_evidence.json
+
+# Cognee local graph DB / logs (created next to the package)
+.cognee/
+.cognee_system/
+*.cognee_lock
+
+# ─── logs ─────────────────────────────────────────────────────────────
+logs/
+*.log
+
+# ─── os / editor ──────────────────────────────────────────────────────
+.DS_Store
+Thumbs.db
+ehthumbs.db
+desktop.ini
+
+.idea/
+.vscode/
+!.vscode/extensions.json
+!.vscode/settings.shared.json
+*.swp
+*.swo
+*~
+
+# ─── misc ─────────────────────────────────────────────────────────────
+tmp/
+.tmp/
+.scratch/

cogneemind/.gitignore

@@ -0,0 +1,84 @@
+# ─── secrets ──────────────────────────────────────────────────────────
+# NEVER commit any env or credential file. Track .env.example instead.
+.env
+.env.*
+!.env.example
+*.pem
+*.key
+*.crt
+secrets/
+credentials.json
+service-account*.json
+
+# ─── python ───────────────────────────────────────────────────────────
+.venv/
+venv/
+env/
+__pycache__/
+*.py[cod]
+*$py.class
+*.egg
+*.egg-info/
+.eggs/
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+.coverage
+.coverage.*
+htmlcov/
+build/
+develop-eggs/
+.installed.cfg
+.tox/
+.nox/
+pip-log.txt
+pip-delete-this-directory.txt
+
+# ─── node / vite / frontend build ─────────────────────────────────────
+node_modules/
+dist/
+dist-ssr/
+.vite/
+.cache/
+.parcel-cache/
+.turbo/
+*.tsbuildinfo
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+pnpm-debug.log*
+.pnpm-store/
+
+# ─── runtime / generated data ─────────────────────────────────────────
+# Wiki + evidence are regenerated at runtime — keep them out of git.
+backend/data/wiki*.json
+backend/data/evidence.json
+backend/data/wiki_evidence.json
+
+# Cognee local graph DB / logs (created next to the package)
+.cognee/
+.cognee_system/
+*.cognee_lock
+
+# ─── logs ─────────────────────────────────────────────────────────────
+logs/
+*.log
+
+# ─── os / editor ──────────────────────────────────────────────────────
+.DS_Store
+Thumbs.db
+ehthumbs.db
+desktop.ini
+
+.idea/
+.vscode/
+!.vscode/extensions.json
+!.vscode/settings.shared.json
+*.swp
+*.swo
+*~
+
+# ─── misc ─────────────────────────────────────────────────────────────
+tmp/
+.tmp/
+.scratch/

cogneemind/README.md

@@ -0,0 +1,116 @@
+# CogneeMind — Airport Gate-Ops Brain
+
+A self-improving Company Brain (Cognee Cloud Hackathon) that assigns airport
+gates **and discovers its own assignment algorithm**, getting better the more
+disruptions it sees. Three agents — **Researcher / Planner / Observer** — share
+an LLM wiki; the Researcher warm-starts each search from what the wiki already
+learned, so discovery **compounds**: fewer evaluations and better plans over time.
+
+> Lower score = better plan. `Score = 1000·U + 500·C + 1·W + 5·R`
+> (U=no-gate, C=conflicts, W=walking, R=reassignments)
+
+![ops deck](frontend/smoke.png)
+
+## What's inside
+
+```
+backend/
+  simulator.py    # gate assignment (greedy by strategy) + scoring
+  disruptions.py  # add_flight / close_gate / apply_delays + signatures
+  agents.py       # Researcher (warm-start discovery) / Planner / Observer + optimize()
+  memory.py       # LocalWiki: recall priors, distill winners + dead-ends, lint
+  brain.py        # Cognee integration: ingest, SkillRunEntry propose->apply, distill
+  api.py          # FastAPI + WebSocket (/scenario, /chat, /compounding, /events)
+  evidence.py     # prints + writes the before/after evidence
+  test_cognee.py  # standalone proof of the Cognee loop (ingest->improve->recall)
+  my_skills/      # researcher | planner | observer | linter  (SKILL.md each)
+  data/           # gates.json, flights.json, rules.md
+frontend/
+  src/Gantt3D.tsx       # three.js / r3f Gantt (glowing bars over a grid deck)
+  src/GanttFallback.tsx # 2D Gantt if WebGL is unavailable
+  src/GanttStage.tsx    # WebGL detection + error boundary -> 3D or 2D
+  src/App.tsx           # ops-deck UI: readout, agent deck, wiki timeline, chat
+```
+
+## Run it (two terminals)
+
+**Terminal 1 — backend** (Python 3.10–3.14). First time only:
+```bash
+cd backend
+uv venv && uv pip install --python .venv/bin/python fastapi "uvicorn[standard]"
+# optional but recommended — turns on the Cognee brain:
+uv pip install --python .venv/bin/python cognee==1.2.0.dev1
+```
+Add your key to `backend/.env` (auto-loaded; gitignored):
+```
+LLM_API_KEY=sk-...
+```
+Then start it (wait ~25–30s for `cognee ready — ingested rulebook + 4 skills`):
+```bash
+.venv/bin/python -m uvicorn api:app --port 8077
+```
+
+**Terminal 2 — frontend**:
+```bash
+cd frontend && npm install && npm run dev
+```
+Open the URL Vite prints (e.g. **http://localhost:5173**).
+
+> No `.env`/Cognee? It still runs **fully** on the local wiki — boots instantly,
+> same demo. Cognee just adds the durable graph + skill self-rewrite on top.
+
+## Demo flow
+Click **Base → +Flight → Close G3 → Storm** and watch the **Agent Deck** narrate
+the loop (Observer flags chaos → Researcher discovers → Planner heals) while the
+Gantt heals to green. Then hit **run cold vs warm** in the Compounding card.
+
+| Step | Score | Point at |
+|------|-------|----------|
+| Base | 6 (U=0) | calm opening plan |
+| + Flight | 70 (U=0) | naive would strand it; brain heals |
+| Close G3 | 31 (U=0) | reuses learned strategy, no re-search |
+| Storm | 36 (U=0) | bars shift in time, heals again |
+| **cold vs warm** | **1003 → 6** | same storm, same 2-try budget — memory wins |
+
+Type in the **Ops Console**: `close G3`, `storm`, `add flight 09:15`,
+`what is the best strategy?` (the last shows a live `[cognee recall: …]`).
+
+## Evidence (no server needed)
+```bash
+cd backend && python evidence.py     # -> data/evidence.json
+```
+- **Self-improvement:** storm under naive `earliest_arrival` strands a flight
+  (**1003**) → discovered `latest_departure` (**6**, all gated).
+- **Compounding:** same storm, 2-try budget — cold **1003** vs warm **6**;
+  full-budget cold needs **6** evals to reach what warm finds in **2**.
+
+## Cognee — verified working
+With `LLM_API_KEY` set, the full judged loop runs against a local
+cognee 1.2.0.dev1 graph (verified live):
+- **Ingest** rulebook + 4 skills (`content_type="skills"`).
+- **Agentic search** reasons over skills + graph (`SearchType.AGENTIC_COMPLETION`).
+- **Self-improve** — a stranded-flight failure records `SkillRunEntry(success=0.0)`;
+  a proposal is generated and **applied** via `improve_skill(apply=True)`
+  (`{'applied': True, 'skill': 'researcher'}`).
+- **Distill + recall** — winners → permanent graph (no `session_id`); raw trials →
+  session memory (`session_id=run-…`). `cognee.recall(...)` returns the learned strategy.
+
+Prove it standalone: `cd backend && .venv/bin/python test_cognee.py`.
+
+## How it maps to the hackathon
+- **Ingest** — rules + skills into the brain.
+- **Query + self-improve** — score each run → `SkillRunEntry` proposes a skill
+  rewrite → `improve_skill(apply=True)`.
+- **Lint** — dedupe/prune/conflict-resolve the wiki (`/lint`).
+- **Two tiers** — raw trials in session memory; winning strategies +
+  disruption→strategy index + dead-ends distilled into the permanent graph.
+
+See `SUBMISSION.md` for the filled-in template and reproducible numbers.
+
+## Troubleshooting
+- **`Could not set lock on file … Lock is held by PID`** — only one backend can
+  use the local Cognee graph at a time. A previous server is still alive:
+  `pkill -9 -f "uvicorn api:app"`, then start again.
+- **Blank 3D area** — WebGL unavailable; it auto-falls back to the 2D Gantt.
+- **`recomputing…` stuck / brain not ready** — backend still ingesting (~30s on
+  boot) or not started; check Terminal 1.