pulse-os 10.14.0

PULSE: A Neural Operating System for AI Swarms.

PULSE OS

Persistent AI Memory for Developer Workflows

Website · Install · Docs · Pricing

---

The Problem

Your AI agents forget everything between sessions. Claude helped you debug a tricky race condition on Tuesday — on Thursday, a different agent walks into the same trap. Every session starts from zero.

PULSE gives your entire AI swarm a shared, persistent brain.

---

What It Does

PULSE runs alongside your existing tools. Every agent session — Claude Code, Gemini CLI, custom workers — automatically captures lessons, failures, and decisions into a local brain. Future sessions boot with that knowledge pre-loaded.

pip install pulse-os
pulse init
pulse activate PULSE-SOLO-cus_xxx-xxxxxxxxxxxxxxxxxxxxxxxx
pulse start
pulse status

  License:       SOLO (trialing) | trial ends 2026-03-15

  Orchestrator:  RUNNING
  Daemons:       7 alive

  Brain Evidence: 1,247 items
  Brain Health:  83 wants | 61 dont_wants | 24 anti-patterns
  Knowledge:     47 lessons | 29 fails | 18 patterns | 8 decisions

---

Key Capabilities

Persistent Memory Every session feeds the brain. Lessons, failures, patterns, decisions — captured automatically, deduplicated, and searchable in ~125ms across 10 parallel sources.

Multi-Agent Swarm One command launches parallel workers across Claude, Gemini, OpenAI, and Ollama. Agents claim tasks from a shared board, execute, and report back without stepping on each other.

pulse swarm                    # Interactive: pick provider → model → count
pulse swarm gemini             # Launch Gemini workers
pulse swarm stop               # Kill all workers

Constitutional Governance Agents operate under a hash-verified Constitution they cannot override. If the Constitution is tampered with, the system refuses to start.

Knowledge Graph Causal relationships between concepts — CAUSES, PREVENTS, REQUIRES, LEADS_TO. The Prediction Engine queries the graph before task execution and injects failure-risk warnings into agent boot.

Claude Code Native PULSE hooks into Claude Code automatically. Every Edit and Write fires a brain query before execution — relevant context injected without any manual prompting.

---

Architecture

pulse/
├── core/           Brain I/O, crypto, LLM routing, paths
├── brain/          Search (10 sources), extraction, save, boot briefing
├── daemons/        Orchestrator + bridge + neural + consolidation + synthesis
├── workers/        Headless task execution, autopilot (CLI swarm)
├── tasks/          Task board, dispatcher, relay
├── graph/          Knowledge graph build + causal query
├── api/            REST endpoints (FastAPI)
└── cli/            47 commands

Your brain lives at ~/.pulse/projects/<hash>/ — local, private, never leaves your machine.

---

Swarm Models

Provider	Models
Anthropic	claude-haiku-4-5 · claude-sonnet-4-6 · claude-opus-4-6
Google	gemini-2.5-flash · gemini-2.5-pro · gemini-3-flash-preview · gemini-3-pro-preview · gemini-3.1-pro-preview
OpenAI	o4-mini · gpt-5.2-codex · gpt-5.3-codex · gpt-5.1-codex-max · o3
Ollama	Any pulled model + cloud-routed: qwen3-coder-next · devstral-2 · glm-5 · minimax-m2.5 · kimi-k2.5

---

Plans

	Solo — $14.97/mo	Pro — $29.97/mo
Brain pipeline	Full	Full
Swarm workers	Up to 5	Unlimited
CLI commands	47	47
Web dashboard	✓	✓
Knowledge graph	View	Full RAG + Causal
Synthesis daemons	—	✓
Docker + self-host	—	✓
Trial	7 days free	7 days free

Start your free trial →

---

How Memory Works

Agent session
    ↓
Capture daemon intercepts conversation
    ↓
Bridge extracts: lessons · failures · patterns · decisions
    ↓
Brain stores in Hippocampus/ (local, cryptographically signed)
    ↓
Consolidation daemon deduplicates + mines patterns every 6h
    ↓
Next session: boot briefing injects relevant context
    ↓
Agent starts smarter than the last one

---

License

Proprietary. Copyright © 2026 PULSE Contributors. All rights reserved.

Brain data stays 100% local — we never access or train on it.

[e-mail]