cameron.stream
A React Native app for the ultimate thinking partner.
Update README and add GitHub link to login screen
- Completely rewrote README: removed emoji, focused on core concept of co as thinking partner
- Added open source link to login screen (links to github.com/letta-ai/co)
- Fixed Refresh Co error handling: catch 404 errors and continue with agent recreation
🐾 Generated with [Letta Code](https://letta.com)
Co-Authored-By: Letta <noreply@letta.com>
+204
-168
+19
-8
App.new.tsx
+19
-8
App.new.tsx
···
386
386
// Delete primary agent
387
387
await lettaApi.deleteAgent(coAgent.id);
388
388
console.log('Deleted primary agent:', coAgent.id);
389
-
390
-
// Clear agent and messages from store
391
-
clearAgent();
392
-
clearMessages();
393
-
// useAgent hook will automatically reinitialize when it detects no agent
394
389
} catch (error: any) {
395
-
if (Platform.OS === 'web') {
396
-
window.alert('Failed to refresh agent: ' + error.message);
390
+
// If agent not found, it's already deleted - continue with refresh
391
+
const isNotFound = error.message?.toLowerCase().includes('not found') ||
392
+
error.status === 404 ||
393
+
error.statusCode === 404;
394
+
395
+
if (isNotFound) {
396
+
console.log('Agent already deleted, continuing with refresh');
397
397
} else {
398
-
Alert.alert('Error', 'Failed to refresh agent: ' + error.message);
398
+
// Real error - show to user and abort
399
+
if (Platform.OS === 'web') {
400
+
window.alert('Failed to refresh agent: ' + error.message);
401
+
} else {
402
+
Alert.alert('Error', 'Failed to refresh agent: ' + error.message);
403
+
}
404
+
return;
399
405
}
400
406
}
407
+
408
+
// Clear agent and messages from store (runs even if agent was already deleted)
409
+
clearAgent();
410
+
clearMessages();
411
+
// useAgent hook will automatically reinitialize when it detects no agent
401
412
};
402
413
403
414
// Load data when view changes
+28
CoLoginScreen.tsx
+28
CoLoginScreen.tsx
···
8
8
ActivityIndicator,
9
9
SafeAreaView,
10
10
useColorScheme,
11
+
Linking,
11
12
} from 'react-native';
12
13
import { StatusBar } from 'expo-status-bar';
13
14
import { useFonts, Lexend_400Regular, Lexend_600SemiBold, Lexend_700Bold } from '@expo-google-fonts/lexend';
···
81
82
<Text style={styles.helpText}>
82
83
Don't have an API key? Visit letta.com to create one.
83
84
</Text>
85
+
86
+
<View style={styles.githubLink}>
87
+
<Text style={styles.githubText}>
88
+
co is{' '}
89
+
<Text
90
+
style={styles.githubLinkText}
91
+
onPress={() => Linking.openURL('https://github.com/letta-ai/co')}
92
+
>
93
+
open source
94
+
</Text>
95
+
{' '}and welcomes contributions
96
+
</Text>
97
+
</View>
84
98
</View>
85
99
</View>
86
100
<StatusBar style="auto" />
···
165
179
color: darkTheme.colors.text.tertiary,
166
180
textAlign: 'center',
167
181
lineHeight: 20,
182
+
},
183
+
githubLink: {
184
+
marginTop: 24,
185
+
alignItems: 'center',
186
+
},
187
+
githubText: {
188
+
fontSize: 13,
189
+
fontFamily: 'Lexend_400Regular',
190
+
color: darkTheme.colors.text.tertiary,
191
+
textAlign: 'center',
192
+
},
193
+
githubLinkText: {
194
+
color: darkTheme.colors.interactive.primary,
195
+
textDecorationLine: 'underline',
168
196
},
169
197
});
+157
-160
README.md
+157
-160
README.md
···
1
1
# co
2
2
3
-
A minimalist chat interface for Letta AI agents. Each user gets their own persistent co agent that learns and remembers across conversations.
3
+
A thinking partner that learns how you think.
4
+
5
+
co is a single AI agent that builds itself around you through conversation. Unlike traditional chatbots, co maintains a persistent, evolving understanding of who you are, what you're working on, and how you approach problems.
6
+
7
+
## Core Concept
8
+
9
+
Every user gets one co agent. That's it. No switching between assistants, no configuring multiple bots. Just one agent that:
10
+
11
+
- **Learns your patterns**: Notices how you think, what excites you, what frustrates you
12
+
- **Remembers everything**: Maintains both short-term context and long-term memory
13
+
- **Evolves with you**: Updates its understanding as you grow and change
14
+
- **Thinks alongside you**: Not just responding, but actively making connections and surfacing insights
15
+
16
+
The entire agent definition lives in `src/utils/coAgent.ts`. When you log in for the first time, co creates itself and begins learning.
17
+
18
+
## How It Works
19
+
20
+
### Single Agent Architecture
21
+
22
+
When you log in, the app:
23
+
1. Looks for an agent tagged with `co-app`
24
+
2. If found, connects to your existing co
25
+
3. If not found, creates a new co agent
26
+
27
+
This means your co persists across devices and sessions. The same agent, with all its accumulated knowledge about you, is available wherever you log in.
28
+
29
+
### Memory System
30
+
31
+
co maintains structured memory across multiple blocks:
32
+
33
+
**You** - Dynamic snapshot of what you're focused on right now. Updated frequently as your attention shifts.
34
+
35
+
**Human** - Stable facts about you. Professional context, communication style, interests. Changes slowly over time.
36
+
37
+
**Tasks** - What co is helping you accomplish. Added as new work emerges, removed when complete.
38
+
39
+
**Knowledge Structure** - How you think. Mental models you use, patterns in how you approach problems, connections between concepts.
40
+
41
+
**Interaction Log** - Significant moments in your relationship. Decisions made, insights gained, understanding that shifted.
42
+
43
+
**Persona** - co's core identity and directives. How it thinks about its role as your thinking partner.
44
+
45
+
**Memory Management** - Guidelines for when to create new memory blocks dynamically based on your usage patterns.
46
+
47
+
### Background Processing
48
+
49
+
co includes a sleeptime agent that runs in the background:
50
+
- Manages long-term archival memory
51
+
- Surfaces relevant context from past conversations
52
+
- Operates independently to maintain the primary agent's performance
53
+
54
+
The sleeptime agent shares memory blocks with the primary agent, enabling seamless context passing.
4
55
5
56
## Features
6
57
7
-
- 🤖 **Single Agent**: One co agent per user, automatically created on first login
8
-
- 🧠 **Persistent Memory**: Advanced memory blocks that evolve over time
9
-
- 💬 **Smooth Streaming**: Token-buffered streaming (50 FPS) for consistent text appearance
10
-
- 🎨 **Polished UI**: Clean, minimal interface with animated message transitions
11
-
- 🌓 **Theme Toggle**: Switch between light and dark modes
12
-
- 🔧 **Tool Support**: Web search, archival memory, conversation search
13
-
- 📱 **Cross-platform**: iOS, Android, and Web support via React Native + Expo
14
-
- 🔒 **Secure**: API token storage with AsyncStorage/SecureStore
58
+
### Persistent Memory
59
+
Every conversation builds on what co already knows. The agent's understanding compounds over time.
15
60
16
-
## Architecture
61
+
### Real-time Streaming
62
+
Messages stream token-by-token on web, creating natural conversation flow.
17
63
18
-
co uses a simplified single-agent architecture:
64
+
### Adaptive Interface
65
+
- Light and dark themes
66
+
- Cross-platform support (iOS, Android, Web)
67
+
- Responsive layout for mobile and desktop
19
68
20
-
1. **Login**: User enters Letta API key
21
-
2. **Agent Discovery**: App searches for agent with `co-app` tag
22
-
3. **Auto-creation**: If no co agent exists, creates one automatically
23
-
4. **Chat**: User chats directly with their co agent
69
+
### File Sharing
70
+
Attach files to messages for context. Works across all platforms.
24
71
25
-
### co Agent Configuration
72
+
### Memory Inspection
73
+
View co's memory blocks to see how its understanding has evolved. Available through the sidebar navigation.
26
74
27
-
co is created with:
28
-
- **Model**: `anthropic/claude-sonnet-4-5-20250929`
29
-
- **Tools**: `send_message`, `archival_memory_insert`, `archival_memory_search`, `conversation_search`, `web_search`, `fetch_webpage`
30
-
- **Memory Blocks**:
31
-
- `persona`: co's adaptive personality
32
-
- `human`: User profile that evolves
33
-
- `approach`: Conversation and memory approach
34
-
- `working_theories`: Active theories about the user
35
-
- `notes_to_self`: Reminders for future reference
36
-
- `active_questions`: Questions to explore
37
-
- `conversation_summary`: Ongoing conversation overview
75
+
### Tool Access
76
+
co can search the web, manage archival memory, and search conversation history to provide informed responses.
38
77
39
78
## Getting Started
40
79
41
80
### Prerequisites
42
81
43
-
- Node.js 18+
44
-
- npm or yarn
45
-
- Expo CLI
82
+
- Node.js 18 or higher
46
83
- Letta API token from [letta.com](https://letta.com)
47
84
48
85
### Installation
49
86
50
87
```bash
51
-
# Install dependencies
52
88
npm install
53
-
54
-
# Start development server
55
89
npm start
56
-
57
-
# For production performance (recommended)
58
-
npm run web:prod
59
90
```
60
91
61
-
### Run Options
92
+
Press `w` to open in browser, or scan the QR code with Expo Go on your mobile device.
62
93
63
-
- **Web**: `npm run web` or press `w`
64
-
- **iOS**: Press `i` (requires Xcode)
65
-
- **Android**: Press `a` (requires Android Studio)
66
-
- **Mobile Device**: Use Expo Go app and scan QR code
94
+
### First Conversation
67
95
68
-
### First Use
96
+
1. Enter your Letta API key
97
+
2. Wait for co to initialize (creates agent with memory blocks)
98
+
3. Start talking
69
99
70
-
1. Launch the app
71
-
2. Enter your Letta API key
72
-
3. Wait for co to initialize (creates agent if needed)
73
-
4. Start chatting!
100
+
co begins building its understanding immediately. The more you interact, the better it gets at thinking alongside you.
101
+
102
+
## Configuration
103
+
104
+
### Agent Definition
74
105
75
-
## Project Structure
106
+
The entire agent is defined in `src/utils/coAgent.ts`. Key configuration:
76
107
77
-
```
78
-
co/
79
-
├── App.tsx # Main application
80
-
├── CoLoginScreen.tsx # Login/authentication
81
-
├── web-styles.css # Web-specific CSS for focus states and themes
82
-
├── src/
83
-
│ ├── api/
84
-
│ │ └── lettaApi.ts # Letta API client
85
-
│ ├── components/
86
-
│ │ ├── MessageContent.tsx # Markdown message rendering
87
-
│ │ ├── ExpandableMessageContent.tsx # Collapsible long messages
88
-
│ │ ├── ToolCallItem.tsx # Tool execution display
89
-
│ │ └── LogoLoader.tsx # Loading animations
90
-
│ ├── types/
91
-
│ │ └── letta.ts # TypeScript definitions
92
-
│ ├── utils/
93
-
│ │ ├── coAgent.ts # co agent creation logic
94
-
│ │ └── storage.ts # AsyncStorage wrapper
95
-
│ └── theme/
96
-
│ ├── index.ts # Theme system
97
-
│ ├── colors.ts # Color palette
98
-
│ └── typography.ts # Font definitions
108
+
```typescript
109
+
model: 'anthropic/claude-haiku-4-5-20251001'
110
+
tools: ['conversation_search', 'web_search', 'fetch_webpage', 'memory']
111
+
enableSleeptime: true
99
112
```
100
113
101
-
## Key Features
114
+
### Memory Blocks
102
115
103
-
### Smooth Token Streaming
116
+
Default memory structure is defined in `src/constants/memoryBlocks.ts`. Each block includes:
117
+
- Label (unique identifier)
118
+
- Description (guides when/how co should update it)
119
+
- Initial value (template structure)
120
+
- Size limit
104
121
105
-
Messages stream at 50 FPS with a token buffer that releases 1-3 characters at a time for consistent, natural text appearance. A hollow circle indicator (○) appears at the end of streaming text.
122
+
co can dynamically create new blocks based on usage patterns (e.g., creating a "decision_log" if you frequently discuss choices, or "technical_notes" if you share code regularly).
106
123
107
-
### Animated Message Layout
124
+
### System Prompt
108
125
109
-
When you send a message:
110
-
1. Message appears at the bottom
111
-
2. An animated spacer grows beneath it (400ms animation)
112
-
3. Your message smoothly rises to the top of the viewport
113
-
4. co's response fills the reserved space below
126
+
co's core behavior is defined in `src/constants/systemPrompt.ts`. This establishes:
127
+
- How co thinks about its role
128
+
- When to update memory
129
+
- How to decompose and synthesize information
130
+
- Relationship dynamics with the user
114
131
115
-
This creates a clean reading experience where your message stays visible with room for the response.
132
+
## Architecture
116
133
117
-
### Theme Support
134
+
Built with React Native and Expo for true cross-platform support:
118
135
119
-
Toggle between light and dark modes with inverted text input styling:
120
-
- **Dark mode**: White background input with black text
121
-
- **Light mode**: Black background input with white text
136
+
- **State**: Zustand stores for agent, auth, and chat state
137
+
- **API**: Letta SDK client with streaming support
138
+
- **Storage**: AsyncStorage (mobile) and SecureStore (sensitive data)
139
+
- **UI**: Native components with custom animations
122
140
123
-
### Memory Viewer
141
+
Key files:
124
142
125
-
Access co's memory blocks through the sidebar to see what co has learned about you and how it's evolving its understanding over time.
143
+
```
144
+
src/
145
+
utils/coAgent.ts - Agent creation and initialization
146
+
constants/memoryBlocks.ts - Memory block templates
147
+
constants/systemPrompt.ts - Agent behavior definition
148
+
hooks/useAgent.ts - Agent lifecycle management
149
+
hooks/useMessageStream.ts - Streaming message handler
150
+
stores/agentStore.ts - Agent state management
151
+
```
126
152
127
-
## Customizing co
153
+
## Development
128
154
129
-
### Modify Personality
155
+
### Running Locally
130
156
131
-
Edit `src/utils/coAgent.ts` and update:
132
-
- System prompt
133
-
- Memory block initial values
134
-
- Available tools
135
-
- Model selection
157
+
```bash
158
+
# Web with production optimizations
159
+
npm run web:prod
136
160
137
-
### Add Memory Blocks
161
+
# iOS
162
+
npm run ios
138
163
139
-
Add new blocks to the `memoryBlocks` array in `createCoAgent()`:
164
+
# Android
165
+
npm run android
140
166
141
-
```typescript
142
-
{
143
-
label: 'custom_block',
144
-
value: 'Custom content here...',
145
-
}
167
+
# Clear cache if needed
168
+
npx expo start -c
146
169
```
147
170
148
-
### Change Model
171
+
### Platform Notes
149
172
150
-
Update the `model` field in `findOrCreateCo()`:
173
+
**Web**: Full feature support including token streaming
151
174
152
-
```typescript
153
-
model: 'openai/gpt-4.1', // or other supported models
154
-
```
175
+
**iOS/Android**: No streaming (SDK limitation), but all other features work
155
176
156
-
## Development
177
+
**Image uploads**: Currently disabled due to SDK validation issues
157
178
158
-
### Tech Stack
179
+
## Deployment
159
180
160
-
- **React Native** + **Expo**: Cross-platform framework
161
-
- **TypeScript**: Type safety
162
-
- **Letta SDK** (`@letta-ai/letta-client`): AI agent framework
163
-
- **AsyncStorage/SecureStore**: Persistent storage
164
-
- **React Native Markdown Display**: Markdown rendering
165
-
- **Lexend Font**: Custom typography
181
+
The app is deployed to Vercel:
166
182
167
-
### Available Scripts
183
+
```bash
184
+
npx expo export --platform web
185
+
npx vercel dist --prod --yes
186
+
```
168
187
169
-
- `npm start` - Start Expo dev server
170
-
- `npm run web` - Run in browser (dev mode)
171
-
- `npm run android` - Run on Android
172
-
- `npm run ios` - Run on iOS
173
-
- `npx expo start -c` - Clear cache and restart
188
+
Production URL will be provided upon deployment.
174
189
175
-
### Production Build
190
+
## Customization
176
191
177
-
For better performance:
192
+
### Changing the Model
178
193
179
-
```bash
180
-
# Development with production optimizations
181
-
npx expo start --web --no-dev --minify
194
+
Edit `src/utils/coAgent.ts`:
182
195
183
-
# Static production build
184
-
npx expo export:web
185
-
npx serve web-build
196
+
```typescript
197
+
model: 'anthropic/claude-sonnet-4-5-20250929' // or any supported model
186
198
```
187
199
188
-
## API Integration
200
+
### Adding Memory Blocks
189
201
190
-
co connects to Letta's API:
202
+
Create new block definitions in `src/constants/memoryBlocks.ts` and add to `getDefaultMemoryBlocks()`.
191
203
192
-
- `GET /agents?tags=co-app` - Find co agent
193
-
- `POST /agents` - Create co agent
194
-
- `GET /agents/{id}/messages` - Load message history
195
-
- `POST /agents/{id}/messages/streaming` - Stream messages
196
-
- `GET /agents/{id}/blocks` - View memory blocks
204
+
### Modifying Behavior
197
205
198
-
## Troubleshooting
206
+
Edit `src/constants/systemPrompt.ts` to change how co thinks about its role and responsibilities.
199
207
200
-
### Agent Not Found
208
+
## Known Limitations
201
209
202
-
If co fails to initialize:
203
-
1. Check API token validity at letta.com
204
-
2. Verify network connection
205
-
3. Check console logs for errors
206
-
4. Try logging out and back in
210
+
- Image uploads disabled (SDK bug with streaming endpoint)
211
+
- Token streaming only works on web (React Native SDK limitation)
212
+
- Agent initialization with sleeptime can take 30-60 seconds
213
+
- "Refresh Co" button may timeout during recreation
207
214
208
-
### Memory Blocks Not Loading
215
+
## Philosophy
209
216
210
-
- Ensure agent is fully initialized
211
-
- Check that `listAgentBlocks()` has proper permissions
212
-
- Verify agent ID is correct
217
+
Most AI assistants are stateless or have shallow memory. They respond to prompts but don't build deep models of who you are.
213
218
214
-
### Streaming Issues
219
+
co is different. It's designed to become your thinking partner through sustained interaction. It notices patterns in how you work, remembers what matters to you, and proactively makes connections you might miss.
215
220
216
-
- Check network stability
217
-
- Verify streaming endpoint support
218
-
- Review console logs for chunk errors
219
-
- Try clearing Expo cache: `npx expo start -c`
221
+
The goal isn't to replace your thinking. It's to augment it. To be the cognitive extension that helps you see your own ideas more clearly.
220
222
221
-
### Slow Performance
222
-
223
-
Run the app with production optimizations:
224
-
```bash
225
-
npx expo start --web --no-dev --minify
226
-
```
227
-
228
-
## License
229
-
230
-
MIT License
223
+
One agent. One relationship. Built over time through conversation.
231
224
232
225
## Resources
233
226
234
227
- [Letta Documentation](https://docs.letta.com)
235
-
- [Letta LLMs.txt](https://docs.letta.com/llms.txt)
228
+
- [Letta Discord](https://discord.gg/letta)
236
229
- [React Native Docs](https://reactnative.dev)
237
230
- [Expo Docs](https://docs.expo.dev)
231
+
232
+
## License
233
+
234
+
MIT