Implement Synix v0.1: build system for agent memory

pipeline.py

@@ -0,0 +1,73 @@
+# pipeline.py — Personal Memory Pipeline (Monthly Rollups)
+# This is the default pipeline for the demo
+
+from synix import Pipeline, Layer, Projection
+
+pipeline = Pipeline("personal-memory")
+
+pipeline.source_dir = "./exports"
+pipeline.build_dir = "./build"
+
+pipeline.llm_config = {
+    "model": "claude-sonnet-4-20250514",
+    "temperature": 0.3,
+    "max_tokens": 1024,
+}
+
+# --- Layers ---
+
+# Level 0: Raw conversation transcripts
+pipeline.add_layer(Layer(
+    name="transcripts",
+    level=0,
+    transform="parse",
+))
+
+# Level 1: Episode summaries (one per conversation)
+pipeline.add_layer(Layer(
+    name="episodes",
+    level=1,
+    depends_on=["transcripts"],
+    transform="episode_summary",
+    grouping="by_conversation",
+))
+
+# Level 2: Monthly rollups
+pipeline.add_layer(Layer(
+    name="monthly",
+    level=2,
+    depends_on=["episodes"],
+    transform="monthly_rollup",
+    grouping="by_month",
+))
+
+# Level 3: Core agent memory
+pipeline.add_layer(Layer(
+    name="core",
+    level=3,
+    depends_on=["monthly"],
+    transform="core_synthesis",
+    grouping="single",
+    context_budget=10000,
+))
+
+# --- Projections ---
+
+# Search index — hierarchical search with provenance
+pipeline.add_projection(Projection(
+    name="memory-index",
+    projection_type="search_index",
+    sources=[
+        {"layer": "episodes", "search": ["fulltext"]},
+        {"layer": "monthly", "search": ["fulltext"]},
+        {"layer": "core", "search": ["fulltext"]},
+    ],
+))
+
+# Context document — ready-to-use agent system prompt
+pipeline.add_projection(Projection(
+    name="context-doc",
+    projection_type="flat_file",
+    sources=[{"layer": "core"}],
+    config={"output_path": "./build/context.md"},
+))

pipeline_topical.py

@@ -0,0 +1,86 @@
+# pipeline_topical.py — Personal Memory Pipeline (Topic-Based Rollups)
+# Alternate pipeline for the demo "config change" moment
+# Same data, different architecture — episodes grouped by topic instead of month
+
+from synix import Pipeline, Layer, Projection
+
+pipeline = Pipeline("personal-memory-topical")
+
+pipeline.source_dir = "./exports"
+pipeline.build_dir = "./build"
+
+pipeline.llm_config = {
+    "model": "claude-sonnet-4-20250514",
+    "temperature": 0.3,
+    "max_tokens": 1024,
+}
+
+# --- Layers ---
+
+# Level 0: Same raw transcripts (CACHED from first run)
+pipeline.add_layer(Layer(
+    name="transcripts",
+    level=0,
+    transform="parse",
+))
+
+# Level 1: Same episode summaries (CACHED from first run)
+pipeline.add_layer(Layer(
+    name="episodes",
+    level=1,
+    depends_on=["transcripts"],
+    transform="episode_summary",
+    grouping="by_conversation",
+))
+
+# Level 2: CHANGED — Topic-based rollups instead of monthly
+# LLM clusters episodes by topic, then synthesizes each cluster
+pipeline.add_layer(Layer(
+    name="topics",
+    level=2,
+    depends_on=["episodes"],
+    transform="topical_rollup",
+    grouping="by_topic",
+    config={
+        "topics": [
+            "career-and-job-search",
+            "technical-projects",
+            "san-francisco",
+            "personal-growth",
+            "ai-and-agents",
+        ],
+        # At runtime, the topical rollup transform queries the episode
+        # search index (already built) to find relevant episodes per topic.
+        # This scales to 1000+ conversations without blowing up context.
+    },
+))
+
+# Level 3: Core memory — same transform, but new inputs (topics instead of monthly)
+# Will REBUILD because its dependency changed
+pipeline.add_layer(Layer(
+    name="core",
+    level=3,
+    depends_on=["topics"],
+    transform="core_synthesis",
+    grouping="single",
+    context_budget=10000,
+))
+
+# --- Projections ---
+
+pipeline.add_projection(Projection(
+    name="memory-index",
+    projection_type="search_index",
+    sources=[
+        {"layer": "episodes", "search": ["fulltext"]},
+        {"layer": "topics", "search": ["fulltext"]},
+        {"layer": "core", "search": ["fulltext"]},
+    ],
+))
+
+pipeline.add_projection(Projection(
+    name="context-doc",
+    projection_type="flat_file",
+    sources=[{"layer": "core"}],
+    config={"output_path": "./build/context.md"},
+))

pyproject.toml

@@ -1,38 +1,20 @@
 [project]
 name = "synix"
 version = "0.1.0"
-description = "Neural schematic primitives for agent memory"
+description = "A build system for agent memory"
 readme = "README.md"
-requires-python = ">=3.12"
+requires-python = ">=3.11"
 license = "MIT"
 authors = [
     { name = "Mark", email = "mark@example.com" }
 ]
-keywords = ["agent", "memory", "cognitive-architecture", "llm"]
-classifiers = [
-    "Development Status :: 2 - Pre-Alpha",
-    "Intended Audience :: Developers",
-    "License :: OSI Approved :: MIT License",
-    "Programming Language :: Python :: 3.12",
-    "Topic :: Scientific/Engineering :: Artificial Intelligence",
-]
+keywords = ["agent", "memory", "build-system", "llm"]
 
 dependencies = [
     "click>=8.0",
     "pydantic>=2.0",
-    "pydantic-settings>=2.0",
     "rich>=13.0",
-    "sqlalchemy>=2.0",
-    "httpx>=0.25",
-    "jinja2>=3.0",
-]
-
-[project.optional-dependencies]
-dev = [
-    "pytest>=8.0",
-    "pytest-asyncio>=0.23",
-    "ruff>=0.3",
-    "pyright>=1.1",
+    "anthropic>=0.40.0",
 ]
 
 [project.scripts]
@@ -42,13 +24,19 @@ synix = "synix.cli:main"
 requires = ["hatchling"]
 build-backend = "hatchling.build"
 
+[dependency-groups]
+dev = [
+    "pytest>=8.0",
+    "pytest-asyncio>=0.23",
+    "pytest-cov>=5.0",
+]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+
 [tool.ruff]
 line-length = 100
-target-version = "py312"
+target-version = "py311"
 
 [tool.ruff.lint]
 select = ["E", "F", "I", "UP", "B", "SIM"]
-
-[tool.pyright]
-pythonVersion = "3.12"
-typeCheckingMode = "strict"

sprint-checklist.md

@@ -0,0 +1,92 @@
+# Synix Sprint Reference Card
+## Fri-Sun Feb 7-9, 2026
+
+### Friday Night (NOW — 2 hrs)
+- [ ] Fill in YC app text fields
+- [ ] Record 1-min founder video (iPhone, one take)
+- [ ] Submit YC app with GitHub repo link
+- [ ] Set up Synix repo with CLAUDE.md + pyproject.toml
+- [ ] `uv init` + `uv sync`
+- [ ] Prepare test data: full ChatGPT + Claude exports in ./exports/
+
+### Saturday (main build day — 8 hrs)
+**Morning (Phase 1-2: Foundation + Parsers)**
+- [ ] Artifact store + tests
+- [ ] Provenance + tests
+- [ ] ChatGPT parser + tests
+- [ ] Claude parser + tests
+- [ ] Transform base class + prompt template loading
+- [ ] `uv run pytest` — all green before continuing
+
+**Afternoon (Phase 3: Pipeline Engine)**
+- [ ] Config parser + tests
+- [ ] DAG resolver + tests
+- [ ] Pipeline runner + tests (mock LLM)
+- [ ] LLM transforms (episode, monthly, topical, core) + tests
+- [ ] `uv run pytest` — all green before continuing
+
+**Evening (Phase 4: Projections + CLI)**
+- [ ] Search index projection + tests
+- [ ] Flat file projection + tests
+- [ ] CLI commands (run, search, lineage, status) + tests
+- [ ] Rich formatting (progress bars, colored output, tree views)
+- [ ] `uv run pytest` — all green before continuing
+
+### Sunday (integration + demo — 6 hrs)
+**Morning**
+- [ ] Integration tests (pipeline run, incremental rebuild, config change)
+- [ ] Run on full dataset (1000+ conversations)
+- [ ] Fix bugs, re-run tests
+- [ ] `uv run pytest` — ALL tests green
+
+**Afternoon**
+- [ ] Run test_demo_flow.py — the automated demo sequence
+- [ ] Demo rehearsal — manual run-through for recording
+- [ ] Screen record demo video
+- [ ] Upload to YC app
+
+### Monday (buffer — 3 hrs)
+- [ ] Bug fixes if needed
+- [ ] Re-record if needed
+- [ ] Final submission before 8pm PT
+
+---
+
+### Budget
+- LLM API: ~$15-30 (Sonnet for all transforms, full dataset)
+- 1000+ convos × episode summary = ~1000 LLM calls (layer 1)
+- ~12 monthly rollups or ~5 topical rollups (layer 2)
+- 1 core synthesis (layer 3)
+- Second run (config change): only layer 2+3 rebuild = ~6-10 calls
+
+### Fallback Demo
+If CLI polish isn't clean enough, record a Python REPL session:
+```python
+from synix.pipeline.runner import run
+from synix.projections.search_index import SearchIndex
+
+result = run("pipeline.py", "./exports")
+# show: 50 built, 0 cached, projections materialized
+
+index = SearchIndex("./build/search.db")
+results = index.query("anthropic")
+# show: results with provenance chains
+
+# show the context doc projection
+print(open("./build/context.md").read()[:500])
+# show: agent system prompt output
+
+# change config...
+result2 = run("pipeline_topical.py", "./exports")
+# show: 0 transcripts rebuilt (cached), 0 episodes rebuilt (cached),
+#        8 topic rollups built, 1 core rebuilt, projections re-materialized
+```
+
+### Scope Fence — DO NOT BUILD
+❌ Stateful artifacts / scratchpad
+❌ Pipeline branching / A/B testing
+❌ Postgres / Neo4j / any external DB
+❌ Web UI / API server
+❌ Eval harness / cost estimation
+❌ Docker / deployment
+❌ Anything not needed for the 90-second demo