SemeSeam Documentation
Complete guide to state condensation, small, named spines, character-exact quote checks, and autonomous descents.
Getting Started
Run the carrel from source, launch Chat Lab, connect local Hermes or an API key, and run your first study session.
Core Concepts
The mathematical and structural foundations: spine bundles, the four spine kinds, lattices, Greek markup, and descents.
Ways to Connect
Architecture options for driving reasoning: local engine, BYO keys, X subscription, MCP connectors, or external skills.
Privacy Architecture
Exact boundaries: what stays on your computer, what leaves over utility pipes, and the free-tier data warning.
Quick Start Walkthrough
Run SemeSeam Carrel
Today SemeSeam runs from source: start the carrel server at http://127.0.0.1:8791. Packaged standalone desktop builds are planned (see the Download Page for sample channels).
Choose a Reasoning Engine
Open Settings → Models:
- Local Hermes 3 (Zero Cost): Run
ollama pull hermes3:8b. The carrel automatically detects local Ollama instances and runs on your hardware. - Sign in with X: Built-in device-code flow, currently off pending an issued client ID from xAI to SemeSeam (or users can provide their own).
- Bring Your Own Key: Enter an API key from Anthropic, Google, or OpenAI (keys are stored on your machine in
data/keys.jsonwith 0600 permissions, not encrypted). Set recorded monthly caps in Settings → Spending.
Pose a Problem in Chat Lab
In Chat Lab, attach research files or pose a prompt. When referencing documents, SemeSeam's Find and Quote algorithm searches the actual files on disk, ensuring every citation is verbatim before displaying the green verified seal (✓).
Harvest Settled Meaning (Spine Bundle)
When a section of work reaches conclusions, click Save bundle. SemeSeam condenses the dialogue into settled decisions, walls, context, and open steps. The conversation continues with a Fresh Context: the old tokens are discarded, keeping the model sharp and preventing attention degradation.