Skip to content

getting-started.md

Latest commit

 

History

History
117 lines (79 loc) · 4.13 KB

getting-started.md

File metadata and controls

117 lines (79 loc) · 4.13 KB

Getting Started

Build your first memory pipeline in 5 minutes.

What you'll build

A pipeline that takes source documents, produces episode summaries, rolls them up by month, and creates a searchable core memory document. Every insight traces back to its source.

Prerequisites

Step 1: Create a project

uvx synix init my-project --template 01-chatbot-export-synthesis
cd my-project

This creates:

  • sources/ — where your source data goes
  • pipeline_monthly.py — a working pipeline definition
  • .env.example — template for your API key

Copy the env template and add your key:

cp .env.example .env
# Edit .env: ANTHROPIC_API_KEY=sk-ant-...

Step 2: Add source data

Drop your data into sources/. The template includes a sample session transcript to start with.

Sources can be anything — conversation exports (ChatGPT, Claude JSON), documents, reports, notes, markdown, plain text. Synix detects common formats automatically.

Step 3: Build

uvx synix build pipeline_monthly.py

What happens:

  1. Sources parsed — files are read from ./sources/
  2. Episodes generated — each source gets an episode summary (1:1, via EpisodeSummary)
  3. Monthly rollups — episodes are grouped by month and synthesized (N:M, via MonthlyRollup)
  4. Core memory — all rollups are compressed into a single core memory document (N:1, via CoreSynthesis)

Each step only runs if its inputs or prompts changed. Re-run synix build after adding new sources — only the new episodes process; everything else is cached.

Step 4: Explore

uvx synix list                          # all artifacts, grouped by layer
uvx synix show episode-2024-03-15       # render one episode
uvx synix lineage episode-2024-03-15    # trace it back to its source

The lineage command is the provenance chain — you can see exactly which source produced each episode, which episodes produced each monthly rollup, and which rollups produced the core memory.

Step 5: Release and search

Build produces immutable artifacts. Release materializes them into a searchable index:

uvx synix release HEAD --to local
uvx synix search "how did I feel about the move" --release local

Results show matching artifacts with scores. Add --trace to see provenance for each result:

uvx synix search "how did I feel about the move" --release local --trace

Step 6: Change something

Open pipeline_monthly.py and edit the episode summarization prompt, or change MonthlyRollup to TopicalRollup to cluster by theme instead of calendar month.

uvx synix build pipeline_monthly.py

Only downstream artifacts rebuild. If you changed the episode prompt, episodes regenerate but the source parsing is cached. If you added a new source file, only the new episode processes — existing episodes and their downstream rollups stay cached.

Step 7: Use it from your agent

After releasing, your agent can query memory at inference time:

import synix

project = synix.open_project("./my-project")
mem = project.release("local")

# Search memory
results = mem.search("return policy", mode="hybrid", limit=5)
for r in results:
    print(f"[{r.layer}] {r.label} ({r.score:.2f})")
    print(f"  {r.content[:200]}...")

# Or load the core memory as flat context
context = mem.flat_file("context-doc")
# → inject into your agent's system prompt

Synix runs offline — it processes sources into structured memory. Your agent reads the output at runtime. They're decoupled.

What's next

  • Customize the pipeline: Pipeline API — all transform types, projections, validators
  • Integrate with your agent: Integration Guide — SDK, MCP server, CLI, direct file access
  • Understand the architecture: Architecture — the four-tier memory model
  • Try other templates: uvx synix init --list-templates to see all available templates
  • Learn about caching: Cache Semantics — how Synix decides what to rebuild