Getting Started

Your first 5 minutes with Kandela — from install to your first recalled memory.

Installation

macOS / Linux

curl -fsSL https://kandela.ai/install | bash

Windows

Open Git Bash (installed with Git for Windows) and run:

curl -fsSL https://api.kandela.ai/install -o /tmp/ki.sh && bash /tmp/ki.sh

Already have WSL? You can also run the macOS/Linux command inside WSL.

Requirements

  • Claude Code installed and working (claude --version)
  • Python 3.8+ (used internally by the installer)
  • Git for Windows (Windows only — provides Git Bash)

What gets installed

  • MCP server connection (14 memory tools)
  • 8 lifecycle hooks (auto-recall, auto-save, prompt guard, etc.)
  • 22 slash commands (/kd-*)

To uninstall completely: curl -fsSL https://kandela.ai/uninstall | bash

After installation, Kandela hooks into Claude Code automatically. The next time you start a session, it will be active.

Scenario 1: Your First Session

Open your project directory and start Claude Code as usual.

cd ~/projects/my-web-app
claude

Initialize Kandela for this project:

> /kd-init my_web_app

Kandela creates a memory project called my_web_app and adds a small config block to your CLAUDE.md. Memories for this project will now persist across sessions.

Save your first memory

Kandela's hooks automatically save important context in the background. You can also ask Claude to store something explicitly:

> We decided to use PostgreSQL instead of SQLite
  because we need concurrent write support.
  Please remember this.

What to store (and what not to)

Worth storing

  • Why you chose PostgreSQL over SQLite
  • Gotcha: "migrations must run before seeding"
  • Deploy command: docker compose up -d
  • API key location: .env has STRIPE_KEY

Skip these

  • File structure or import paths
  • Function signatures
  • Code patterns Claude can read
  • Things obvious from the repo

Scenario 2: Next Session — Auto Recall

Close Claude Code. Come back tomorrow.

cd ~/projects/my-web-app
claude

Kandela's Auto Recall kicks in automatically:

[Kandela] Brief recall loaded (3 memories)
- DB: PostgreSQL chosen for concurrent writes
- Gotcha: run migrations before seed
- Deploy: docker compose -f prod.yaml up -d

You didn't ask for this — it just appears. Claude now has the context it needs from previous sessions, right from the start.

Need more detail mid-conversation?

> What was the database migration issue we ran into last week?

Claude uses Kandela's context_search to look it up. No special syntax — just ask.

Scenario 3: Searching Past Memories

After a few weeks, you'll have dozens of memories. Kandela's hybrid search finds what you need.

> Search my memories for anything about deployment.
[1] Deploy: use --no-deps flag to avoid restarting other containers (imp: 9.0)
[2] Staging server at staging.myapp.com, SSH port 2222 (imp: 5.0)
[3] CI/CD: GitHub Actions, deploys on merge to main (imp: 6.0)

Cross-project search

> Search all my projects for how I configured Redis caching.

Finds relevant memories even if they were stored in a different project.

Scenario 4: Remote Control via Telegram

Connect the Telegram bot once, then manage your projects from anywhere — commute, meetings, or bed.

Connect in 30 seconds

Go to Dashboard → Account → Telegram Link, scan the QR code with Telegram, and you're connected.

Telegram features require a Pro plan or higher.

Status check

/status

See active projects, memory counts, recent sessions, and server health — all at a glance.

Save a memo

Remember: checkout flow needs
to handle expired sessions

Jot down ideas on the go — a bug you just noticed, an architecture thought during lunch, or a task for tomorrow. Next session, Kandela delivers it via /kd-inbox.

Switch projects

/project my_backend

Switch the active project context. Subsequent memos and searches target the selected project.

Search memories

/search redis caching config

Search across your memories right from Telegram. Results show matching memories with importance scores.

Daily summary

/daily

Get today's work summary — what was accomplished, decisions made, and pending items. Great for end-of-day review.

Remote command

/do @my_web_app run the failing
tests and fix them

Your agent picks up the command, executes it in the project workspace, and sends results back to Telegram. Use it for running tests, fixing builds, or deploying updates — without opening your laptop.

PIN confirmation for risky commands

/do @prod_server deploy to production

High-risk commands (production deploys, database migrations, destructive operations) trigger a PIN confirmation step. You'll see a confirmation prompt with details — enter your PIN within 60 seconds to proceed.

Scenario 5: Direct Control

When you want to be explicit about what gets remembered.

Store with specific importance

> Store this as importance 9: "Never run migrations on prod
  without a backup. We lost 2 hours on March 3rd."

High-importance memories always appear in Auto Recall.

Check project status

> /kd-status

Shows current project, memory count, recent sessions, and inbox items.

Load another project's memories

> /kd-load my_infrastructure

Cross-reference deployment rules, API keys, or shared gotchas from another project.

Save a gotcha

> Remember this as a gotcha: "docker compose up without
  --no-deps restarts ALL containers. Always use --no-deps."

Next time a deploy command runs, Prompt Guard surfaces this warning.

Update or delete

> Search for memories about the staging server.
> Update that memory — staging moved to staging-v2.myapp.com.
> Delete the memory about the old staging server.

Summarize the session

> /kd-update

Saves a session summary and checks for unsaved decisions. Good habit at the end of a productive session.

Cheat Sheet

CommandWhat it does
/kd-init <id>Initialize Kandela for a new project
/kd-statusShow current project status
/kd-inboxReview unread memos
/kd-listList all your memory projects
/kd-load <name>Load another project's memories
/kd-updateSave session summary + sync
/kd-syncSync cached memories to server
/kd-progressFull project progress report
/kd-helpShow all available commands

What Happens Automatically

You don't need to manage most features manually:

Session start

Recalls relevant memories from past sessions

During work

Monitors context and saves important decisions

On failures

Records gotchas so the same mistake isn't repeated

Session end

Summarizes and persists what happened

Context compression

Preserves key info before the context window compacts

Next Steps