Agentic AI Inside Your Chat: How Ditto’s Sub-Agents Research, Create, and Act
Everyone’s talking about AI agents. The pitch is always the same: give the AI a goal and let it figure out the steps. Sounds great in a demo. In practice, most AI assistants still work one prompt at a time — you ask, they answer, you ask again. There’s no delegation, no parallel execution, no multi-step workflows happening while you keep working.
And when an AI does run tools or agents behind the scenes? You usually have no idea what it’s doing until it’s done. Black box in, result out. Hope it did the right thing.
Ditto takes a fundamentally different approach. Sub-agents are built into the conversation — they run inside your threads, stream progress in real-time, and ask for approval before taking costly actions. You see everything. You control everything. And your agents benefit from the same persistent memory that powers every Ditto conversation.
What Can Ditto’s Agents Actually Do?
Ditto ships with specialized agents that extend what the AI can do beyond text generation:
Research Agent
Ask Ditto to research a topic and it doesn’t just generate text from training data. It searches the web, reads pages, follows links, and synthesizes findings — all within your current thread.
"Research the latest changes to the Web Audio API
and summarize what's new for developers."Ditto spawns a research sub-agent that:
- Searches the web for relevant sources
- Reads and extracts content from the top results
- Synthesizes a summary with citations
- Returns the findings directly into your conversation
The results are saved as a memory, so next time you ask about Web Audio, Ditto already has the research to build on. No re-Googling. No repeating yourself.
Art Agent
Describe an image and Ditto generates it using state-of-the-art models. The generated image is stored alongside the conversation context — so weeks later, you can reference “that logo concept we created” and Ditto knows exactly what you mean.
"Generate a minimal logo for a developer tool called Patchwork —
think clean geometry, dark background, subtle gradient."Because images live in your knowledge graph, you can iterate across sessions. Come back next week and say “try the Patchwork logo again but with warmer tones” — the AI has the full creative history.
Web Search
When a conversation needs current information — stock prices, documentation changes, recent events — Ditto can search the web and pull in live results without you leaving the chat. This isn’t a separate “browse” mode. It’s a tool the AI invokes naturally when your question calls for it.
The Difference: Agents Run Inside Your Threads
Here’s what makes Ditto’s approach different from “AI agents” you’ve seen elsewhere.
In most AI products, agent execution happens in a separate context — a background process, a dedicated “agent mode,” or a sandboxed environment that produces a final result. The agent doesn’t have access to your conversation history, your preferences, or your ongoing work.
In Ditto, sub-agents run inside your threads. That means they inherit:
- Your memory — the sub-agent can search your personal knowledge graph and use past conversations as context
- Your thread context — attached subjects, memories, and notes are visible to the agent
- Your goals — if you’ve set active goals, the agent factors them into its work
When a sub-agent finishes, its results are folded back into the thread’s conversation history. The work the agent did doesn’t vanish into a log file somewhere — it becomes part of your persistent, searchable memory.
Real-Time Visibility: No Black Boxes
This is the part most agentic AI products get wrong. They hide what agents are doing until they’re done. You wait. You hope. You get a result and have no idea how the AI got there.
Ditto streams sub-agent progress in real-time. Each sub-agent gets its own SSE stream that runs concurrently within your thread. You can see:
- What tool was called — research, image generation, web search
- What arguments were used — the search query, the generation prompt
- Intermediate results — as they arrive, not just the final output
- Which thread spawned it — background indicators show when non-active threads have agents running
If you’re in a different thread and an agent completes in the background, you’ll see an indicator without losing your current context. Multitasking works the way it should.
Guardrails: Agents Ask Before Acting
Not all actions are free — image generation costs tokens, and some tool executions have real consequences. Ditto implements an approval system for high-impact actions.
When a sub-agent needs to do something that costs money or has side effects:
- The agent pauses and shows you what it wants to do
- You see the estimated cost in USD before anything happens
- You choose: approve, deny, or revise the instruction
- The agent resumes with your explicit decision
This isn’t just a safety feature. It’s a trust feature. You stay in control of what your AI does on your behalf, and the AI never burns tokens on something you didn’t want.
Agents + Memory = Compounding Intelligence
The real power of Ditto’s agentic system isn’t any single agent — it’s how agents interact with your memory over time.
Here’s a realistic example:
Week 1: You ask Ditto’s research agent to investigate state management patterns for your SolidJS app. It searches, reads, and synthesizes a comparison. That research is saved as a memory.
Week 2: You’re in a different thread, debugging a reactivity issue. Ditto’s memory retrieval pulls in the state management research from last week — automatically — because it’s semantically relevant. The AI references specific patterns from that earlier research.
Week 3: You ask for help picking between two approaches. The AI draws on both the original research and the debugging context to give you a recommendation grounded in your actual project history.
No copy-pasting from browser tabs. No “remember that thing I researched?” The agents did the work, memory preserved it, and future conversations build on it.
How to Use Ditto’s Agents
You don’t need to configure anything. Just talk to Ditto naturally:
- “Research X for me” — triggers the research agent
- “Generate an image of X” — triggers the art agent
- “What’s the latest on X?” — may trigger web search if current info is needed
Agents are available on every model, in every thread. They respect your thread context, your goals, and your memory.
Try It
Head to assistant.heyditto.ai and ask Ditto to research something you’ve been meaning to look into. Watch the agent work in real-time. Come back tomorrow and see how that research shows up in a completely different conversation.
That’s the difference between an AI that runs tools and an AI that learns from them.