Memory That Thinks

Proactive Context Layer

AI agents forget as context windows overflow. Sulcus captures raw session signals, filters out conversational noise, and consolidates verified facts, preferences, and procedures into permanent working memory.

CLI Installer$ curl -fsSL https://sulcus.ca/install.sh | bash
Start Free
View SDKs
View Plans
Explore Flow
COREAUTO CAPTURESIGNAL LOCKWORKING MEMORYHOT CONTEXTPROACTIVE RECALLSUB 25MSSMART SYNTHESISZERO BLOATCLAUDE CODECURSOR IDECODEX CLIOPENCLAW
Autonomous Memory

Proactive Context Engine

Passive vector stores wait until an agent remembers to query them. Sulcus operates automatically around every turn: relevant context flows in before your agent speaks, and newly discovered facts consolidate when the turn completes.

  • Automatic Recall — Surfaces relevant preferences, decisions, and procedures in under 25 milliseconds before every response.
  • Continuous Learning — Extracts new insights and engineering fixes in the background with zero prompt bloat.
  • Connected Knowledge — Links people, repositories, and decisions across your entire agent fleet.
  • Smart Synthesis — Condenses older episodic history into high-salience summaries, cutting token usage by 90 percent.

Memory Lattice

Send a high-salience harmonic wave across the memory lattice.
Automatically rotate focus across all four active memory nodes.
Core
Signal Extract
Auto Capture

Ingest raw session streams and extract structured memory traces automatically.

Sub 25ms
Working Memory

Assemble active context across the right memory node before each turn.

Graph Match
Proactive Recall

Surface resonant memories from the bottom golden index automatically.

Zero Bloat
Smart Synthesis

Condense episodic history into durable high-salience summaries.

Ingest raw session streams and extract structured memory traces automatically.
Assemble active context across the right memory node before each turn.
Surface resonant memories from the bottom golden index automatically.
Condense episodic history into durable high-salience summaries.

Auto Capture:Raw agent streams enter the top node and split into structured memory traces.

Signal Lock:99.4% Accuracy
Live Recall

Instant Grounded Answers

Launch Web
Instant Context
ClaudeMCPCursor
Agent Prompt

“What froze the Cerberus authentication schema?”

Working Context

ADR-041 locked the Keycloak token claims format on October 14. Two microservice namespaces still depend on legacy realm roles before migration completes.

FACT · ADR-041PROCEDURAL · cerberus-jwtSYNTHESIS · auth-rollout
Core Architecture

Built For Scale

Four autonomous memory stages that keep enterprise agent context accurate, fast, and self-cleaning.

01

Intelligent Extraction

Every interaction passes through automated extraction. Facts, preferences, and procedures persist while conversational noise is filtered out.

02

Connected Knowledge

Memories link across engineers, repositories, and harnesses. When an agent recalls a topic, it receives the complete constellation of context.

03

Adaptive Retention

Thermodynamic salience keeps critical instructions permanently hot while ephemeral debugging chatter cools according to archetype decay curves.

04

Memory Compression

Historical episodes condense into verified synthesis nodes in the background, preserving full continuity at a fraction of the token budget.

System Specification

Under The Hood

01

Thermodynamic Salience

Memories carry active salience heat, stability, and archetype decay curves so critical knowledge stays hot while transient noise cools.

02

Golden Index

Stratifies memory across preference, fact, procedural, semantic, episodic, and synthesis layers with verified and corroborated trust.

03

Fast Recall

Assembles token-budgeted working context in under 25 milliseconds using hybrid vector similarity and graph resonance spread.

04

Reactive Triggers

Automated event rules on store, recall, decay, and boost pin preferences, reinforce recalled skills, and dispatch webhooks.

05

Local Binary

Runs 100 percent airgapped on your machine with embedded PostgreSQL and CPU embeddings, or syncs selectively to Sulcus Cloud.

06

Namespace Governance

Two-Gate Namespace ACLs, Cerberus JWT authentication, and prefix-glob rules isolate and govern multi-agent fleet access.

07

Universal Harnesses

Connects Claude Code, Cursor, Codex, OpenClaw, Python, Node.js, and MCP so knowledge never dies in a single terminal.

Engine Parameters

Complete Control

Tune context budgets, resonance spread, reinforcement gains, and retention half-lives through clean API parameters.

Active Index

max_nodes
context_budget
hot_threshold
cold_threshold

Graph Resonance

spread_factor
depth
damping
filter_gate

Memory Reinforcement

on_recall
on_update
on_edge
stability_gain

Smart Consolidation

cold_trigger
cold_threshold
strategy
batch_size

Engine Cadence

interval_ms
trigger_ops
max_idle_ms
worker_concurrency

Retention Curves

half_life
floor_salience
reinforce_curve
decay_rate
Event Automation

Reactive Memory Triggers

Attach automated rules directly to your memory graph. Pin critical facts, boost recalled skills, and trigger external webhooks without manual prompting.

on_storepin

Pin Preferences

Pin preference memories automatically the moment they are captured so core instructions never decay.

on_recallboost

Reinforce Recall

Boost salience heat whenever a procedural memory is recalled so high-value skills stay in working context.

on_decaynotify

Cooling Alerts

Emit an event notification when domain knowledge begins cooling down before it leaves active memory.

on_boostwebhook

Fleet Webhooks

Dispatch signed webhooks to your CI pipelines or agent orchestrators when key memories are reinforced.

Developer Experience

Five Line Integration

Install the native CLI binary with a single command, or install our zero-dependency Python and Node.js SDKs directly from the official GitHub repository.

Read Docs

CLI Sidecar

Single-process Rust binary with embedded local storage, CPU embeddings, and native Model Context Protocol server.

Shell Installer
curl -fsSL https://sulcus.ca/install.sh | bash
Homebrew Install
brew tap digitalforgeca/sulcus && brew install sulcus

Python SDK

pip install "git+https://github.com/digitalforgeca/sulcus.git#subdirectory=sdks/python"
from sulcus import Sulcus

s = Sulcus(api_key="sk-...")
s.remember("User prefers dark mode",
           memory_type="preference")
results = s.search("dark mode")

Node.js SDK

npm install "github:digitalforgeca/sulcus#subdirectory=sdks/node"
import { Sulcus } from "@digitalforgestudios/sulcus-nodejs";

const s = new Sulcus({ apiKey: "sk-..." });
await s.remember("User prefers dark mode",
  { memoryType: "preference" });
const results = await s.search("dark mode");
Data Sovereignty

Your Infrastructure Rules

Run the Sulcus binary 100 percent locally by default, and choose selectively which project namespaces sync across your encrypted cloud boundary.

01

Airgapped Binary

Runs entirely on your local hardware with embedded PostgreSQL and local CPU embeddings. Zero outbound network calls required.

02

Selective Sync

Choose which project namespaces remain strictly on your local workstation and which delta-sync to Sulcus Cloud.

03

Managed Keys

Encrypt memory payloads locally before sync using Customer Managed Keys for cryptographic isolation at rest and in transit.

04

Enterprise Fleet

Coordinate concurrent agent teams with Two-Gate Namespace ACLs, prefix-glob governance, and the real-time Fleet Event Bus.

Production Telemetry

Validated Performance

<25ms
Recall Latency
87
API Endpoints
v2.9
Engine Release
15k+
Graph Vertices

Start Building Today

Free tier included. Connect Claude Code, Cursor, Codex, or your custom agent fleet in under five minutes.

Create Account
Read Docs

Engineering Dispatches

Releases, thermodynamic memory research, and architecture updates from Digital Forge Studios.