ActiveLog runs a lightweight model locally that thinks continuously. A supervisor shapes its thoughts over time — building permanent intelligence that compounds.
Today's AI is brilliant and static. Every conversation starts from zero. No memory of yesterday's insights. No evolution from experience.
Each session wipes the slate clean. Hard-won reasoning is lost the moment the context window closes.
Models only think when prompted. There's no background process connecting dots or reflecting on past work.
Training stops at deployment. The model never gets smarter about your specific domain, codebase, or workflow.
ActiveLog pairs a lightweight local model with a supervisor. The thinker runs in a loop — observing, reflecting, connecting. The supervisor shapes its reasoning, curating which thoughts are worth keeping. Over time, the system develops genuine expertise.
A small model runs continuously on your machine, generating a stream of thoughts about whatever it's pointed at.
A shaping layer evaluates, redirects, and reinforces thought patterns — building expertise instead of noise.
Every valuable thought is journaled. The system builds a permanent, searchable record of evolving intelligence.
Switch the thinker's strategy without changing the architecture. Each mode is a lens — a different way of processing the same continuous stream.
Relentlessly questions assumptions. Digs into data, follows leads, and writes up findings like a journalist on deadline.
Argues the opposite of every conclusion. Stress-tests logic, finds holes, and surfaces the weakest links in reasoning.
Reflects ideas back transformed. Takes input and returns variations, inversions, and unexpected connections.
Monitors the web continuously. Tracks changes, detects patterns, and flags what matters across any set of sources.
Finds patterns across disparate inputs. Links ideas that seem unrelated, building a web of understanding over time.
Runs what-if scenarios in a loop. Tests hypotheses, simulates outcomes, and ranks approaches by predicted success.
A clean separation between thinking, shaping, and acting. Each layer is independently swappable.
Install the CLI, point it at a data source, and watch thoughts stream in real-time.
# Install
npm install -g @activelog/cli
# Start a thought stream
activelog init \
--mode investigative \
--source ./my-data \
--supervisor shaping
# Watch it think
activelog stream --liveThinker and supervisor are model-agnostic. Use local Llama, Phi, or any API.
Files, databases, APIs, web pages, or raw event streams — anything that produces text.
Your thoughts never leave your machine. No cloud, no telemetry, no phoning home.
Free and open source. Runs entirely on your machine. No account required.