@agentforge-ai/runtime

@agentforge-ai/runtime

Persistent Mastra runtime package for AgentForge.

What This Package Exports

The public surface is defined in src/index.ts.

Agent Factory And Memory

• createStandardAgent
• initStorage
• getStorage
• getVector
• createStandardMemory
• isStorageInitialized
• runtime defaults: DAEMON_MODEL, OBSERVER_MODEL, EMBEDDING_MODEL, DEFAULT_TOKEN_LIMIT

These helpers implement the repo’s Convex-backed memory pattern using @mastra/convex.

Model Registry

• getModel
• getModelsByProvider
• getActiveModels
• getContextLimit
• resolveModel

Built-In Tools

• datetimeTool
• webSearchTool
• readUrlTool
• manageNotesTool

Daemon And Channels

• AgentForgeDaemon
• channel abstractions: ChannelAdapter, AgentDefinition, DaemonConfig
• channel implementations: HttpChannel, DiscordChannel, TelegramChannel
• shared streaming helpers: progressiveStream, splitMessage, formatSSEChunk, generateThreadId

Current Runtime Shape

In this repo, the runtime package is the home of the persistent daemon architecture. The CLI start command lazily imports this package to instantiate agents, initialize optional Convex-backed memory, and boot the HTTP channel.

Channel status in the current CLI flow:

• HTTP channel: started by agentforge start
• Discord channel: export exists, CLI wiring is present, but the current start command reports it as not yet implemented
• Telegram channel: export exists, CLI wiring is present, but the current start command reports it as not yet implemented

Development

pnpm --filter @agentforge-ai/runtime build
pnpm --filter @agentforge-ai/runtime test
pnpm --filter @agentforge-ai/runtime typecheck

License

Apache-2.0