God Mode
Always-on screen context
Active window, clipboard, and running apps auto-injected into every AI call. BLADE knows what you're working on without being told.
Desktop AI · Open Source · Free
v1.5 just shippedThe AI that watches,
learns, and works.
60+ tools, 24 MCP servers, Ghost Mode, screen memory, coding agents — wired through a v1.5 agentic loop that verifies its own work, recovers from errors, and injects only the context each query needs. Every other desktop AI talks. BLADE acts.
60+ ToolsVerified Loop@anchorsStuck DetectionSysadminGhost Mode5 AgentsTotal Recall24 MCP ServersFree + OSS
60+ Native Tools
Bash, files, browser, display, sysadmin, screen recording, agents.
Ghost Mode
AI overlay invisible to screen share on Windows.
24 MCP Servers
Auto-discovers and installs integrations as you work.
Ghost Mode
You see the answer.
Nobody else sees the overlay.
BLADE listens to your meetings via system audio. Every 5 seconds, it transcribes the last chunk, detects questions aimed at you, and fires AI-generated suggestions into a transparent always-on-top overlay — written in your own voice. On Windows, it is invisible to screen share via content protection.
Zoom · Meet · Teams · Discord · Slack
Alex Chen
You
Sam
Priya
BLADE · Ghost Mode
Blue-green deployment — no downtime. Safe to confirm the rollout.
Invisible to screen share↵ use
BLADE Swarm · complex goal active
ResearchScanning 14 docs
AnalystProcessing results
MemoryContext injected
WriterWaiting on Analyst
ReviewerQueued
Synthesis pass starts when Analyst completes
BLADE Swarm
5 agents in parallel.
Nobody else does this.
Give BLADE a complex goal. It decomposes it into a dependency graph, routes each step to the best model for the job, and runs up to 5 specialized agents simultaneously. Findings from one feed directly into the next. A final synthesis pass combines all results into a single coherent answer.
No other desktop AI runs more than one agent at a time.
Total Recall
Rewind.ai,
open-source and free.
BLADE captures a screenshot every 30 seconds, fingerprints it — identical frames are skipped — and runs a vision model description for semantic search. Ask what you were debugging Tuesday. Get the exact screenshot and context in seconds.
Rewind.ai · $30/moBLADE · free
What error was I debugging Tuesday?
Tue 2:14 PM · VS Code
TypeError: Cannot read properties of undefined (reading 'id')
Tue 2:31 PM · Terminal
Resolved — null check added to fetchUser(), tests passing
Tue 3:05 PM · Vercel
Deployment complete · fix/null-check → prod
BLADE · chat
Compare@file:src/main.rsagainst@memory:project-deadlineusing@screenfor context.
3 anchors injected · 1.2k tokens⏎ send
Mid-loop verifier · on track
3 / 8 tool callsContext budget9.8k / 24k · 41%
memory 28%repo-map 22%tools 18%screen 12%5 gated off
Stuck detector · NoProgress armed · 0/3
v1.5 · Intelligence Layer
Tools are the medium.
The loop is the moat.
Every other AI runs a naive 12-iteration for-loop with everything injected every turn. v1.5 verifies its own progress, recovers from structured errors, detects when it's stuck, fans out to parallel sub-agents on big tasks, and injects only the context each query needs.
- Selective context injection — every section gated by query relevance. Proactive compaction at 80%, tool outputs capped at 4k tokens.
- Verified agentic loop — mid-loop verifier every 3 tool calls; structured ToolError on failure; truncation auto-retries with higher max_tokens.
- Stuck detection + sessions — 5-pattern detector (RepeatedAction, Thrashing, NoProgress, MonologueSpiral, CostRunaway). Reopen-and-resume from any compaction boundary.
- Auto-decomposition — 5+ independent steps fan out to parallel sub-agents in isolated context windows. Only summaries return.
- Context intelligence — tree-sitter symbol graph + personalized PageRank repo map. Type @screen, @file:, or @memory: to inject anchors as chat chips.
26-fixture eval suite. Top-1: 26/26.
And that's not all
Every other AI does one thing.
BLADE does all of it.
Computer Use
60+ tools, full desktop control
Click buttons, fill forms, read UI with OCR, navigate apps, record screen, manage displays. Vision-driven autonomous loop: screenshot → analyze → act → repeat.
System Admin
GPU passthrough? Done.
Deep hardware detection, IOMMU group mapping, dry-run previews for dangerous changes, task checkpoints across reboots with rollback. Cross-platform package management.
Typed Memory
7 semantic categories
Facts, preferences, decisions, relationships, skills, goals, routines — each stored and retrieved differently. Memory that compounds week over week.
Audio Timeline
Always-on conversation memory
Mic and system audio captured, transcribed, and indexed. 'What did we decide an hour ago?' returns the exact transcript and a summary.
Personality Mirror
Writes exactly like you
Analyzes your writing across chat history — formality, humor, signature phrases, greeting patterns. Every reply sounds like you wrote it.
Security Fortress
Network, phishing, secrets
Monitors active connections, scans for exposed keys and credentials, checks URLs for phishing indicators, flags secrets in code.
How it stacks up.
Compared to the closest alternatives on the market.
| Capability | BLADE | Cluely | Jan | Screenpipe | Open Interpreter |
|---|---|---|---|---|---|
| Verified agentic loop (mid-loop verifier) | ✓ | ✗ | ✗ | ✗ | ✗ |
| @file / @screen / @memory anchors | ✓ | ✗ | ✗ | ✗ | ✗ |
| 60+ native tools | ✓ | ✗ | ✗ | ✗ | partial |
| Parallel multi-agent swarms | ✓ | ✗ | ✗ | ✗ | ✗ |
| System admin (GPU, IOMMU, dry-run) | ✓ | ✗ | ✗ | ✗ | ✗ |
| Screen timeline / Total Recall | ✓ | ✗ | ✗ | ✓ | ✗ |
| Ghost Mode (meeting overlay) | ✓ | partial | ✗ | ✗ | ✗ |
| 24 auto-install MCP servers | ✓ | ✗ | ✗ | ✗ | ✗ |
| Computer use (vision loop) | ✓ | ✗ | ✗ | ✗ | partial |
| Free + open source | ✓ | ✗ | ✓ | ✓ | ✓ |
Download Blade · v2.3
Your machine has a brain now.
60+ tools. Verified agentic loop. @screen / @file: / @memory: anchors. System admin. Screen memory. Free, open source, and built to run your desktop — not just answer questions.
Windows · macOS · Linux · Open source · v2.3