- Category
- AI Agents
- Rank
- No. 1133Tools index
Previous survey · No. 1138 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- moeru-ai
- GitHub
- 34 stars
- Latest release
- v0.0.8
- Date
About
Stream-first agent runtime for building real-time AI agents with token-by-token output.
What it does
Apeira is a TypeScript library for coordinating agent turns as sequences of typed lifecycle, text, reasoning, tool-call, and result events. Applications can consume each turn through a readable stream, or subscribe to an agent and submit work without waiting. Optional packages add sessions, storage adapters, protocol bridges, skills, roleplay, and context compaction.
Why it's ranked here
Apeira merits attention for a clear event model and unusually practical extension points. The supplied code shows more than a thin model wrapper: it manages turn state, storage, sessions, tool execution, protocol translation, and history compaction. The main reservation is scope clarity. The quick start is concise, but it demonstrates only one model API path.
What's good
The event vocabulary covers turn starts, completion, failure, reasoning, text, tool calls, and tool results. Plugins can extend tools, transform stored entries, react after turns, initialize resources, and clean them up. The MCP integration caches clients and tool catalogs, handles partial server failures, follows pagination, responds to tool-list changes, and supports progressive discovery. Compaction preserves recent entries and records summaries in storage.
Tradeoffs
This is infrastructure, not a finished assistant or hosted agent service. Consumers must supply model credentials, instructions, event handling, and an application surface. The quick start centers on an OpenAI-compatible Responses endpoint, so the supplied text does not establish broad provider coverage. Several useful capabilities live in separate packages, which adds selection and integration work. Compaction can fall back to hard truncation after repeated failures.
How to use it well
Use Apeira when a TypeScript application needs explicit control over agent events, tools, persisted history, and protocol adapters. It fits custom chat clients, terminal interfaces, or agent backends that already own their product shell. Start with the core package, then add sessions, storage, MCP, AG-UI, or compaction only when needed. Look elsewhere for a turnkey user interface or managed hosting.
Technical notes+
packages/core/src/index.ts exposes the agent, runner, queue, state, storage, plugin, fork, tool, and run surfaces, while packages/apeira/src/index.ts assembles core, chat, Responses, session, and storage exports into the umbrella package. packages/plugin-mcp/src/index.ts normalizes server configuration, lazily caches clients and catalogs, paginates tool discovery, invalidates caches on tool-list notifications, isolates failures with Promise.allSettled, and closes clients and transports during shutdown. packages/plugin-ag-ui/src/index.ts translates Apeira events into AG-UI run, reasoning, text, step, and tool events while maintaining per-turn state. packages/plugin-compact/src/plugin.ts triggers summarization from token usage, stores compact entries, preserves configured recent input, and falls back to a truncation message after repeated failures.
Observed
- License
- MIT
- Primary language
- TypeScript
- Packaging
- Published as a core package and an umbrella package, installed with pnpm.
- Primary interface
- TypeScript library with readable turn streams and subscription-based event delivery.
- Protocol interfaces
- Separate plugins provide MCP client integration and AG-UI event translation.
- Repository structure
- Private pnpm workspace with recursive build and test scripts and multiple feature packages.
Read from README.md, package.json, packages/core/src/chat.ts, packages/core/src/index.ts, packages/apeira/src/index.ts, packages/session/src/index.ts, packages/storage/src/index.ts, packages/core/src/responses.ts, packages/plugin-mcp/src/index.ts, packages/plugin-ag-ui/src/index.ts, packages/plugin-skills/src/index.ts, packages/plugin-compact/src/index.ts, packages/plugin-compact/src/split.ts, packages/plugin-roleplay/src/index.ts, packages/plugin-compact/src/plugin.ts.
What it can do
Build real-time AI agents
Agent configuration and parameters → Functional AI agent runtime
Stream token-by-token AI responses
User queries or prompts → Real-time streaming text output
Process real-time data streams
Live data feeds or streaming inputs → Processed real-time responses
Execute agent workflows
Workflow definitions and trigger events → Automated agent actions and responses
Manage agent runtime environments
Runtime configuration settings → Active agent execution environment
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.