The AI Agent Studio is the conversational interface built into Narakim for building, understanding, and iterating on bots using natural language. Instead of manually wiring every node, you describe what you want and the AI builds the automation for you.
What the AI Studio Does
- Creates bots — Names, configures archetype, and sets up the bot entity.
- Generates automation graphs — Installs the right extensions and writes the full node/edge graph from a description.
- Explains existing automations — Ask "what does this bot do?" and get a plain-English walkthrough.
- Suggests extensions — Recommends the right extensions for your use case.
- Guides configuration — Walks you through required settings and secrets for each extension.
- Deploys bots — Can trigger a deployment to development directly from chat.
- Troubleshoots errors — Paste an error from your log stream and get a diagnosis and fix.
- Validates and smoke-tests — Runs graph validation and smoke tests on your behalf.
Accessing the AI Studio
The AI Studio is accessible from the AI tab in the studio sidebar. This is the global AI interface — it has access to all your bots and can create new ones.
When you're viewing a specific bot, the AI Studio automatically receives that bot's context — its automation graph, installed extensions, platforms, and configured secrets.
Skills
The AI Agent routes your request to one of four specialist skills based on your intent:
| Skill | Best for |
|---|---|
| Builder | Creating bots, building automations, installing extensions, deploying |
| UI Builder | Designing and publishing Custom UIs — dashboards, portals, public pages bound to your graph |
| Debugger | Diagnosing errors, reading logs, fixing broken graphs |
| Advisor | General questions about Narakim, platform capabilities, how-to guidance |
The skill is selected automatically — you don't need to specify it.
Conversation Context
When accessed with an active bot, the AI receives:
- The bot's name, archetype, status, and environment
- The full automation graph (nodes, edges, and current config)
- Installed extensions and their schemas
- Connected platforms and their status
- Configured secret keys (values are never exposed)
This means you can ask context-aware questions like:
"Why isn't my If Condition branching correctly?"
"Add a Log node after the AI node."
"What happens if the HTTP Request fails?"
"Deploy this bot to development."
Building with AI
Starting from Scratch
Open the AI Studio and describe your bot:
"I want a Discord bot that listens for messages containing the word 'help' in the #support channel. When it sees one, it should use AI to generate a helpful response and send it back."
The AI will:
- Create the bot (if one isn't already active).
- Install the required extensions.
- Build the automation graph (nodes, edges, and mappings).
- Tell you which secrets to configure in Configurations → Extension Secrets.
Iterating
Continue the conversation to refine:
"Add error handling — if the AI call fails, log the error and send a fallback message."
"Change the AI model to Claude 3 Haiku."
"Now deploy it to development."
AI Secret Policy
The AI will never ask you to paste credential values into the chat. Instead it:
- References secret key names (e.g.,
DISCORD_BOT_TOKEN) - Instructs you to go to Configurations → Extension Secrets to fill them in
- If a user pastes a secret into chat, the AI will acknowledge it and immediately remind the user to rotate the credential if it was shared elsewhere
Tool Transparency
Every action the AI takes is shown as a tool card in the conversation — you can see exactly what the agent did, expand it to inspect inputs and outputs, and trace the full sequence of operations.
| Tool Card Status | Meaning |
|---|---|
| ⟳ Spinning | Tool is running |
| ✅ Green | Tool succeeded |
| ❌ Red | Tool failed — expand to see the error |
Retry on Error
If the AI encounters an error (e.g., a tool call fails, the model produces a bad response), a Retry pill appears below the failed message. Click it to automatically re-send the last prompt and try again.
Example Conversations
"Build me a weekly report bot"
Use a Timer Trigger set to run every Monday at 9 AM. Connect it to an HTTP Request node to fetch your data. Feed the data into a Transform Data node to format it. Then use another HTTP Request to send the report to Slack.
"How do I access the user's name in a Discord trigger?"
From a Message Create trigger, the user's display name is at {{nodes.trigger.output.author.username}}. Their user ID is at {{nodes.trigger.output.author.id}}.
"My Telegram bot isn't responding to /start"
Make sure you have the Telegram Command trigger installed and configured with the command name start (no slash). Also verify your TELEGRAM_BOT_TOKEN is set in Configurations → Extension Secrets.
"Add Slack notifications to my existing Discord bot"
Since the bot's archetype is Discord, you'd need to change it to Universal (in Configurations → General → Archetype) to enable multi-platform connections. Then connect Slack in Configurations → Connect Platform, and I'll add a Slack Send Message action to the automation.