⬡ v6.10.1 — Grok joins the desktop app

aegiscode

A desktop AI coding app with multi-model support, semantic memory, council voting, MCP, hooks, and cloud sync. Works with Claude, DeepSeek, Groq, OpenAI, Ollama and more. BYOK — your keys, your cost.

Install now → ★ Get Pro €2/mo

Free & open source · MIT · Pro unlocks higher rate limits & priority support · Upgrade →

✦ Native Chat — now the first thing you see

No terminal window, no separate app — model switching, memory search, and slash commands, all in the desktop app's first tab.

What's new in v6.10.1

AEGIS keeps getting sharper. Here's what changed. Full history at /patch-notes.

𝕏

Grok joins the desktop app

xAI's official grok CLI is now a real, installable option in the Terminal picker, alongside AEGIS Code and Claude Code — wired into the same shared-memory bridge the rest of AEGIS uses.

Native Chat is now the first thing you see

The desktop app opens straight into Chat — no terminal window required. Real slash commands too: /status, /memory, /router, /copy, and more.

Memory pricing, spelled out

Cross-session memory is free, always, stored locally on your machine. The only paid tier is optional cross-device sync — Mobility (€2/mo, 50MB) or AEGIS VIP (€19/mo, 2GB).

Sign-in bug fixed

The "Activate" flow after pasting an API key used to hang indefinitely. Fixed — it now completes and closes properly.

Up in 60 seconds

Download the desktop app for your platform.

Linux .deb

dpkg · x64 · Full Electron desktop app

↓ .deb v6.10.2

macOS .dmg

Apple Silicon · drag to Applications

Coming soon

Windows .exe

Windows 10 / 11 · x64

↓ .exe v6.10.2
See the desktop app →

Setting up your API keys

BYOK — bring your own key, pay your model provider directly, no markup. The desktop app won't start a session until at least one provider is configured.

1
Open Settings

In the desktop app's navbar, click ⚙ Settings.

2
Pick a provider

Claude (ANTHROPIC_API_KEY, from console.anthropic.com → API Keys), OpenAI (platform.openai.com → API keys), Gemini (aistudio.google.com → Get API key), Groq (console.groq.com → API Keys), or DeepSeek (platform.deepseek.com → API keys). Running fully local with Ollama needs no key at all.

3
Paste the key and save

It's stored locally on your machine — never sent anywhere except directly to that provider's API.

4
Switch to Terminal or Chat

The setup banner clears once a key is detected, and you can start a session — or use the built-in Chat tab, which works with any OpenAI-compatible key the same way.

Persistent memory across every session

Aegiscode remembers your codebase, decisions, and context — automatically. Here's how it works.

1
Automatic indexing

Every file you open and every prompt you send is embedded into a local semantic index — no manual tagging required.

2
Vector search at your fingertips

Open the Memory tab (⬡ Memory) to search across everything you've worked on — "how did I fix that auth bug last week?"

3
Cross-session continuity

Close the app, come back tomorrow — the AI picks up right where you left off, with full context of your project history.

4
Sync across machines

Free with any account — your memory index is synced to every machine running aegiscode: desktop, laptop, all in sync.

5
Memory MCP server

Connected tools (Claude, Cursor, VS Code) can read and write to your memory via the built-in MCP server — aegiscode memory:mcp.

6
Privacy first

All embeddings are stored locally by default. Sync is end-to-end encrypted. Your code never leaves your machines unless you choose to share.

Cloud sync settings

Keep your semantic memory in sync across every machine running aegiscode — desktop, laptop, homelab server.

1
Activate sync

Log in for free at aegiscloud.org — or run aegis-cli login. No payment required.

2
Link your machine

Open Settings → Cloud in the desktop app, enter your activation code — your local memory index is linked to your account.

3
Auto-sync

Every session is synced automatically when aegiscode closes. The Cloud tab shows last sync status.

4
Check status

The Cloud tab in the Settings panel shows last sync time, device count, and available memory.

🔒 End-to-end encrypted

The full power of ÆGIS in a desktop app

aegiscode-gui is a native Electron app with the full aegiscode AI engine inside. API keys go in a settings panel, not a .env file. Split-pane view: AI terminal on top, real bash shell below.

/ desktop
ready
Workspace
Terminal
History
Memory
Account
Cloud
Settings
v6.9.0
┌─ aegiscode ─ v6.9.0
│ ÆGIS AI Agent · /help for commands
refactor auth middleware to use JWT refresh tokens
◆ Reading auth/middleware.py…
◆ Writing auth/middleware.py · auth/tokens.py…
→ Replaced session-based auth with stateless JWT. Added
   refresh endpoint at /auth/refresh. Expiry: 15 min / 7 day.
Terminal ~/project
neo@machine:~/project$ git diff --stat
 auth/middleware.py | 48 ++++--
 auth/tokens.py     | 72 +++++++

AI + shell, split view

The AI terminal sits above a real interactive bash shell. Ask aegiscode to write code, then run it immediately below — without switching windows.

🔑

Keys in Settings, not .env

Paste your Anthropic, DeepSeek, Groq or Gemini key into the Settings tab. Saved to ~/.aegiscode/.env automatically.

Session history & memory

Browse past sessions and resume any conversation with one click. The Memory tab shows your full semantic index — search and scroll through everything aegiscode remembers.

Download desktop app →

Built for real work

Everything you need to build, decide and remember — all from your desktop.

Multi-model

Switch between Claude, Gemini, OpenAI, DeepSeek, Groq and Ollama mid-session with /model. Each model remembers the same context.

Auto-router

Routes every task to the cheapest model that can handle it — simple edits to fast models, hard reasoning to frontier ones. It learns from each outcome and re-weights routes, so cost keeps dropping while quality holds. Toggle with /router.

Semantic memory

Cross-session vector memory, free with any account — aegiscode indexes every session automatically, no subscription needed. Works with every model including Ollama.

Council voting

Use /council to send questions to Claude, DeepSeek and Llama simultaneously. They deliberate and vote. Consensus wins.

Debate

/debate makes your models argue it out across rounds — each sees the others' case and pushes back. Reasoned answers on hard calls, not one model's guess.

Private AI — no keys

Run fully local with Ollama, no API key needed. A free aegiscloud.org account (one-time Google login) is required to start the app, but nothing you run locally ever leaves your machine. Bring keys only when you want the cloud models.

Orchestration

Delegate to specialized agents — architect, implementer, reviewer, debugger — working in parallel or in sequence.

MCP Server

Extend capabilities via the Model Context Protocol. Connect external tools, databases and APIs as MCP servers.

Hooks

PreToolUse, PostToolUse and more. Write custom hooks in any language to intercept and modify behavior.

Themes

Choose from default, light, dark, ocean, forest and sunset themes. Set via config or AEGIS_THEME env var.

BYOK

Bring your own API keys. You pay your model provider directly. No markup, no middleman.

Cloud sync — free

Log in at aegiscloud.org — no payment needed. Cross-session semantic memory plus every session synced across machines. Access your history anywhere.

Local first

Memory stored locally at ~/.aegiscode/. Cloud sync is opt-in. Your data stays yours.

Document Vision

Upload PDFs, DOCX, or images — ÆGIS extracts every picture, runs OCR, and analyzes diagrams/charts/UI via Grok Vision or GPT-4o. Pipe extracted text straight into your chat context.

Custom Skills

Teach AEGIS new abilities in plain language with /skill-creator — no coding required. Skills are reusable, version-controlled, and loaded automatically per project. Build your own agent toolkit.

Native Chat

A built-in chat tab talks to any OpenAI-compatible endpoint — no separate CLI window needed. Bring your own key and base URL, switch providers per session.

AEGIS VIP

Early access to every new feature before it ships broadly, plus AEGIS Limited unlocked — same account, no separate key. €19/mo.

Memory's free. €2/month for more speed & support.

The desktop app is fully open-source and free — cross-session memory and cloud sync are included for every account, no subscription needed. Paid plans unlock a higher API rate limit, priority support and full export & backup. BYOK — you pay your model provider directly, we never mark up tokens.

Free

€0

Forever. MIT open source.

  • ✓ Full desktop app
  • ✓ All AI models (Claude, DeepSeek, etc.)
  • ✓ /router auto-model selection
  • ✓ MCP & hooks
  • ✓ Semantic memory (context limited)
  • ✓ Cloud sync across devices
  • ✗ 20 API req/min (Pro: 120)
Get started free →
POPULAR

Pro

€2/month

Cancel anytime. No lock-in.

  • ✓ /council · /debate · /multi · /research
  • ✓ 120 API req/min
  • ✓ Priority support
  • ✓ Full export & backup
  • ✓ Everything in Free
VIP

AEGIS VIP

€19/month

Cancel anytime. No lock-in.

  • ✓ Early access — new features first
  • ✓ AEGIS Limited included
  • ✓ Everything in Pro
Subscribe €19/month →

🏫 School & EDU emails get free Pro automatically. Register with .edu email →

🏢 Need team/enterprise? Contact us

Built different. Priced different.

Every competitor locks you to one model or charges per-seat. aegiscode is model-agnostic, BYOK, and costs €0 for core features. Find this anywhere else.

Feature aegiscode Claude Code Cursor Cline
Multi-model ✓ Claude + DS + Groq + Gemini + Ollama ✗ Claude only ✗ Limited models ✓ BYOK
Semantic memory ✓ Cross-session
Council voting ✓ Multi-model consensus
BYOK · pay your provider ✓ No markup ✗ 20% markup ✗ $20/mo + API
Desktop app ✓ Electron + split shell ✗ Standalone CLI ✗ VS Code extension
Price Free · Pro €2/mo $20/mo $20/mo Free
See pricing →

What users say

⭐⭐⭐⭐⭐

"Switched from Claude Code. Being able to use DeepSeek, Groq and Ollama from the same session is a game changer. Memory across sessions actually works."

— Marcus J., Senior Engineer

⭐⭐⭐⭐⭐

"I run Ollama on a headless server. aegiscode gives it persistent memory, free — no other CLI tool does this. Pro's €2/mo is worth it just for /council alone."

— Sarah K., DevOps Lead

⭐⭐⭐⭐⭐

"Council voting caught a subtle bug in my auth logic that Claude alone missed. The debugger agent found the root cause in 30 seconds. Sold."

— Priya R., Full Stack Developer

Join 1,200+ developers using aegiscode daily

Try it now → Share your story →

Ollama with cross-session memory.
Local inference. Zero API cost.

Most Ollama frontends are stateless — every session starts from scratch. aegiscode gives Ollama the same persistent semantic memory as cloud models, using local vector embeddings. Your conversations are indexed, retrieved and injected as context automatically. No API key required.

🔒

Fully private

Inference runs on your machine. Memory stored locally. No tokens leave your device — cloud sync is opt-in, never required.

Permanent context

Vector embeddings index every session. Relevant past context is retrieved automatically before each request — Ollama remembers across weeks.

Auto-compact

When context fills up, aegiscode automatically summarizes and compresses history so long sessions never hit a wall. Or trigger manually with /compact.

Free forever — no credit card

Try the full Ollama + semantic memory experience for free. Every session is indexed automatically and remembered across all future sessions — no subscription needed. See what Pro adds →

# one-time setup — no API key needed
$ ollama pull llama3.2
$ aegiscode-gui # auto-detects Ollama, loads semantic memory
what was that auth bug I fixed last week?
◎ memory · found 3 relevant sessions from 8 days ago
→ You patched a JWT expiry race in auth/middleware.py — token refresh was firing twice on 401...

aegiscode is likely the first desktop AI app to give Ollama cross-session semantic memory.

See ÆGIS in action

Futuristic HUD terminals — real commands, real output, v6.10.1 ÆGIS engine.

⌨ Code generation

⧩ ÆGIS code · v6.9.0
┌─ ÆGIS v6.9.0
│ · /help for commands
add rate limiting to my Express API
Reading src/index.js…
Writing src/middleware/rateLimit.js
Writing src/index.js
→ Added express-rate-limit · 100 req/15 min
   per IP · 429 with Retry-After header
   2 files · +34 lines · 0 errors

⚖ Council voting

⧩ ÆGIS · council
/council postgres vs mongodb for this app?
⬡ Council deliberating…
● Claude — relational data, use Postgres
● DeepSeek — schema fits SQL, agree
● Llama — Postgres, JSONB covers flex fields
✓ Consensus: PostgreSQL (3 / 3)
→ Your data has clear relations and you'll
   need joins. Postgres wins unanimously.

◎ Semantic memory

⧩ ÆGIS · memory
what was that bug I fixed last week?
memory · 3 sessions from 6 days ago
→ You patched a JWT expiry race in
   auth/middleware.py — refresh was
   firing twice on 401, causing double
   token invalidation in prod.
fix the same issue in the mobile API
Writing mobile/auth/middleware.py
   applied same patch · cross-session context

⇄ Multi-model

⧩ ÆGIS · models
model: claude-sonnet-4-6
/model
  1. claude-fable-5
  2. claude-sonnet-4-6 ← current
  3. deepseek-chat
  4. groq-llama
  5. ollama/llama3.2
/model deepseek-chat
Model → DeepSeek Chat
   context and memory carried over

Slash commands reference

All commands are prefixed with /. Type /help in session to see all.

Command
Alias
Description
/help
/? /h
Show all commands
/model
/m
Interactive model switcher
/model <id>
Switch to model by ID
/model list
List all configured models
/model add ...
Add a custom model
/model remove <id>
Remove a model
/clear
/cls
Clear chat history
/compact
Compress context to save tokens
/status
/st
Show session info and token usage
/theme
/t
Switch UI theme
/thinking
Toggle thinking blocks
/copy
/cp
Copy last code block to clipboard
/copy N
Copy Nth code block
/yolo
Toggle auto-approve for all tool calls
/multi <task>
Run task across multiple agents
/council <question>
Multi-model majority vote
/research <question>
Multi-agent research
/memory
Manage semantic memory
/skills
/sk
List loaded skills
/hooks
View and manage hooks
/version
/v
Show version info

One-shot: aegis-cli "refactor this file"

Session management: aegis-cli --continue · aegis-cli --resume <session-id>

Your keys, your models

Switch model anytime with /model — including the brand new Claude Fable 5. All models share the same memory and session context.

Claude Fable 5 NEW

Anthropic · claude-fable-5 · Latest flagship

Claude Sonnet 4.6

Anthropic · claude-sonnet-4-6

Claude Opus 4.8

Anthropic · claude-opus-4-8

Claude Haiku 4.5

Anthropic · claude-haiku-4-5

DeepSeek Chat

DeepSeek · deepseek-chat

DeepSeek Reasoner

DeepSeek · deepseek-reasoner

Llama 3.3 70B

Groq · groq-llama

QwQ-32B (Groq)

Groq · groq-deepseek · reasoning

Gemini 2.5 Pro

Google · gemini-2.5-pro

Gemini 2.5 Flash

Google · gemini-2.5-flash · fast

GPT-5.5 NEW

OpenAI · gpt-5.5

GPT-4o

OpenAI · gpt-4o

o3

OpenAI · o3 · reasoning

Ollama (local)

Free · private · semantic memory included

Any OpenAI-compatible

Custom · Add to config.json

Works with any OpenAI-compatible API

Add your own models to ~/.aegiscode/config.json and switch to them with /model.

~/.aegiscode/config.json
"models": [
  {
    "id": "my-model",
    "name": "My Custom Model",
    "model": "model-name",
    "baseURL": "https://api.myprovider.com/v1",
    "apiKey": "your-api-key"
  }
]

Compatible with Mistral · Together.ai · Cohere · Perplexity · OpenRouter · LM Studio · Ollama and any self-hosted model.

Switch with: /model my-model

Get aegiscode

Desktop app — pick your platform.

Desktop App (aegiscode-gui)

Linux

aegiscode-gui .deb

Full Electron desktop app. AI terminal + shell split view. Installs via dpkg.

Download .deb v6.10.2

macOS

aegiscode-gui .dmg

Apple Silicon build. Drag to Applications and open.

Coming soon

Windows

aegiscode-gui .exe

Windows Installer (NSIS). Works on Windows 10 and 11.

Download .exe v6.10.2

Common questions

Is aegiscode free? +

Yes. aegiscode is free and open-source (MIT). You bring your own API keys and pay your model provider directly — no markup. Cross-session semantic memory and cloud sync are free too, once you log in. €2/month Pro unlocks /multi, /council, /debate, /research and a higher API rate limit. See pricing →

Does aegiscode have a desktop app? +

Yes — aegiscode-gui is a native Electron desktop app with the full aegiscode engine inside. It has a split-pane view: AI terminal on top, real bash shell on the bottom. API keys go in a settings panel — no .env editing. Available for Linux, macOS and Windows. See the desktop app →

Which AI models are supported? +

Claude (Fable 5, Opus 4.8, Sonnet 4.6, Haiku 4.5), DeepSeek Chat & Reasoner, Groq (Llama 3.3, QwQ-32B), Google Gemini 2.5 Pro & Flash, OpenAI (GPT-5.5, GPT-4o, o3), and Ollama for fully local inference. Any OpenAI-compatible API can be added to config.json.

Can it work offline with Ollama? +

Yes. aegiscode supports Ollama for 100% local inference — no API key, no cloud calls. It also gives Ollama cross-session semantic memory using local vector embeddings, so Ollama can remember context across sessions entirely on your machine.

What is council voting? +

/council sends your question to multiple models simultaneously (e.g. Claude, DeepSeek and Llama). They each answer, deliberate, and the consensus wins. Useful for architecture decisions, code review, or any question where a single model's opinion isn't enough.

How is aegiscode different from Claude Code or Cursor? +

aegiscode is model-agnostic and BYOK — you're not locked to one provider. It adds council voting (multi-model consensus), cross-session semantic memory that works even with Ollama, and runs as a native desktop app. Fully open-source (MIT). Claude Code costs $20/mo. Cursor costs $20/mo. aegiscode and its memory are free — €2/mo Pro adds /multi, /council, /debate and /research. Compare plans →

Share aegiscode

/ desktop
ready
Workspace
Terminal
History
Memory
Account
Cloud
Settings
v6.9.0
┌─ aegiscode ─ v6.9.0
│ ÆGIS AI Agent · /help for commands
refactor auth middleware to use JWT refresh tokens
◆ Reading auth/middleware.py…
◆ Writing auth/middleware.py · auth/tokens.py…
→ Replaced session-based auth with stateless JWT.
   Added /auth/refresh. Expiry: 15 min / 7 day.
Terminal ~/project
neo@machine:~/project$ git diff --stat
 auth/middleware.py | 48 ++++--
 auth/tokens.py     | 72 +++++++
⚙️ Skill Creator: Teach me new abilities on the fly — just type /skill-creator and describe the skill in plain language. No coding needed.
Got it