# Token Savings

> Agent Chat uses approximately 50% fewer tokens than Telegram or Discord. Here is why and how it works.

Agent Chat uses approximately **50% fewer tokens** than connecting your agent to Telegram or Discord. This means faster responses and lower costs — without sacrificing quality.

---

## The Numbers

Here is a real comparison from the same agent, same workspace, same tools — only the chat channel is different:

| Metric | Agent Chat | Telegram |
|--------|-----------|----------|
| **Input tokens per message** | ~4,000 | ~63,000 |
| **Context used** | ~20% | ~50% |
| **Response speed** | Faster | Slower |
| **Cost per conversation** | ~50% less | Baseline |

These numbers come from production usage with the same workspace and tools. Your savings will vary based on conversation length and tool usage, but the pattern is consistent.

---

## Why the Difference?

### Telegram / Discord: One Big Thread

When your agent connects to Telegram or Discord, every message goes into a single conversation. Over time, the context grows:

- Message 1: "Check the build status" + response
- Message 2: "Create a GitHub issue" + response
- Message 3: "What is the latest deployment?" + response
- ...
- Message 50: "Review this PR" + all 49 previous messages still in context

By message 50, the agent is processing tens of thousands of tokens of old, irrelevant context just to answer a simple question.

### Agent Chat: Fresh Conversations

With Agent Chat, each conversation starts fresh. When you open a new chat:

- Only the workspace context (available tools, your settings) is loaded
- Only the messages in this specific conversation are included
- No leftover context from other conversations or other team members

When you ask "Review this PR", the agent processes ~4,000 tokens — your workspace context plus this conversation's messages. Not the 50 other things you asked about last week.

---

## How It Adds Up

For a team of 5 people, each having 10 conversations per day:

| | Telegram | Agent Chat |
|---|---|---|
| **Daily input tokens** | ~3.1M | ~1.3M |
| **Monthly input tokens** | ~65M | ~27M |
| **Monthly savings** | — | **~38M tokens** |

The savings grow with team size and usage. More conversations = more savings, because each conversation stays small instead of accumulating context.

---

## Additional Performance Benefits

### Faster Responses

Less input context means the agent processes your message faster. With ~4K tokens instead of ~63K, the agent starts generating a response almost immediately.

### Better Answers

When the context is focused on the current conversation, the agent is less likely to get confused by unrelated previous messages. Fewer distractions = more accurate responses.

### Cache Efficiency

Agent Chat uses prompt caching for the workspace context (available tools, settings). This shared prefix is cached and reused across conversations, further reducing processing time and cost.

---

## When to Use Telegram Instead

Agent Chat does not replace Telegram or Discord for every use case:

| Use Agent Chat when | Use Telegram/Discord when |
|---------------------|--------------------------|
| Working from the dashboard | You need mobile access on the go |
| Team needs separate conversations | You want a shared group discussion |
| Cost and speed matter | You need platform-specific features (buttons, inline bots) |
| You want approvals + activity monitoring | You need notifications pushed to your phone |

Both can connect to the same workspace simultaneously. Use whichever fits the situation.

---

## Next Steps

- [Connect OpenClaw](/agent-chat/connect-openclaw) — get started with Agent Chat
- [Team Access](/agent-chat/team-access) — multiply the savings across your team
- [Approvals](/agent-chat/approvals) — add approval gates at no extra token cost