Lightweight TypeScript agent runtime. Zero heavy dependencies. OpenAI-compatible.
- Agent loop — tool calling, sub-agent delegation, doom-loop detection, context guard
- OpenAI-compatible — works with any
/chat/completionsendpoint (OpenAI, Anthropic, Ollama, etc.) - File-based memory — keyword recall, zero vector DB dependency. Beats Letta/MemGPT on LoCoMo (86% vs 74%)
- Context compaction — automatic fact extraction + LLM summarization when context overflows
- Lifecycle hooks —
before_prompt,before_tool,after_tool,agent_end - Skills — installable SKILL.md extensions with ClawHub (pure JS git clone)
- Channels — Telegram, Cloud IM adapters (auto-detected from env)
- HTTP + WebSocket gateway — zero external dependencies, real-time streaming
- CLI —
lumin agent,lumin serve,lumin health - ~4,900 LOC — single production dependency (Zod)
npm install @prismer/agent-coreimport { runAgent } from '@prismer/agent-core';
process.env.OPENAI_API_KEY = 'sk-...';
await runAgent({
type: 'message',
content: 'What is 2 + 2?',
});# Run agent with a message
lumin agent --message "Hello, world!"
# Start HTTP + WebSocket gateway
lumin serve --port 3001
# Health check
lumin healthconst ws = new WebSocket('ws://localhost:3001/v1/stream');
ws.onopen = () => {
ws.send(JSON.stringify({ type: 'chat.send', content: 'Hello!' }));
};
ws.onmessage = (e) => {
const msg = JSON.parse(e.data);
if (msg.type === 'text.delta') process.stdout.write(msg.delta);
if (msg.type === 'chat.final') console.log('\n---\nDone:', msg.toolsUsed);
};# Chat (synchronous)
curl -X POST http://localhost:3001/v1/chat \
-H 'Content-Type: application/json' \
-d '{"content": "List files in the workspace"}'
# List tools
curl http://localhost:3001/v1/tools
# Health
curl http://localhost:3001/healthAll settings via environment variables. Sensible defaults for standalone use.
| Variable | Description | Default |
|---|---|---|
OPENAI_API_KEY |
LLM provider API key | (required) |
OPENAI_API_BASE_URL |
LLM provider base URL | https://api.openai.com/v1 |
AGENT_DEFAULT_MODEL |
Default model ID | gpt-4o |
WORKSPACE_DIR |
Working directory | ./workspace |
LUMIN_PORT |
HTTP/WS server port | 3001 |
MAX_CONTEXT_CHARS |
Compaction threshold (chars) | 600000 |
MODEL_FALLBACK_CHAIN |
Fallback models (comma-separated) | — |
PRISMER_PLUGIN_PATH |
Path to workspace plugin | — |
TELEGRAM_BOT_TOKEN |
Telegram channel (optional) | — |
LOG_LEVEL |
debug / info / warn / error |
info |
import { createTool } from '@prismer/agent-core';
import { ToolRegistry } from '@prismer/agent-core/tools';
const tools = new ToolRegistry();
tools.register(createTool(
'weather',
'Get current weather for a city',
{
type: 'object',
properties: {
city: { type: 'string', description: 'City name' },
},
required: ['city'],
},
async (args) => {
const res = await fetch(`https://wttr.in/${args.city}?format=j1`);
return JSON.stringify(await res.json());
},
));@prismer/agent-core
├── Core
│ ├── PrismerAgent — agent loop + tool execution + doom-loop detection
│ ├── OpenAICompatibleProvider — LLM client (any /chat/completions endpoint)
│ ├── FallbackProvider — automatic model fallback chain
│ ├── ToolRegistry — tool registration + JSON Schema specs
│ └── EventBus — SSE / WebSocket event streaming
├── Memory
│ ├── MemoryStore (facade) — store / recall / search / recent
│ └── FileMemoryBackend — keyword-based, zero-dependency
├── Infrastructure
│ ├── HTTP + WebSocket server — zero external deps (pure node:http)
│ ├── CLI — agent / serve / health commands
│ ├── SessionStore — session management + compaction state
│ └── Config — Zod-validated, env var override
└── Extensions
├── HookRegistry — before_prompt, before_tool, after_tool, agent_end
├── SkillLoader — SKILL.md + YAML frontmatter
├── AgentRegistry — sub-agent delegation via @mention
└── ChannelManager — Telegram, Cloud IM adapters
import { PrismerAgent } from '@prismer/agent-core/agent';
import { OpenAICompatibleProvider } from '@prismer/agent-core/provider';
import { ToolRegistry } from '@prismer/agent-core/tools';
import { SessionStore } from '@prismer/agent-core/session';
import { MemoryStore, FileMemoryBackend } from '@prismer/agent-core/memory';
import { HookRegistry } from '@prismer/agent-core/hooks';
import { EventBus } from '@prismer/agent-core/sse';
import { loadConfig } from '@prismer/agent-core/config';
import { createLogger } from '@prismer/agent-core/log';
import { VERSION } from '@prismer/agent-core';Zero-dependency file-based memory with keyword recall. Tested on the LoCoMo long-term conversation memory benchmark:
| Model | Overall | No Adversarial | vs Letta/MemGPT |
|---|---|---|---|
| Claude Opus 4.6 | 86% | 95% | +12pp |
| Kimi K2.5 | 63% | 56% | -11pp |
| Letta/MemGPT | ~74% | — | baseline |
Zero-dependency keyword search + strong LLM outperforms Letta's embedding+rerank pipeline.
import { MemoryStore } from '@prismer/agent-core/memory';
const memory = new MemoryStore('./workspace');
await memory.store('The calibration coefficient is 0.03847', ['numeric']);
const results = await memory.search('calibration coefficient');
console.log(results); // [{ content: '...', score: 1.0, tags: ['numeric'] }]Drop markdown files into your workspace to customize agent behavior:
| File | Priority | Purpose |
|---|---|---|
IDENTITY.md / SOUL.md |
10 | Agent identity and persona |
AGENTS.md |
9 | Sub-agent routing and priorities |
TOOLS.md |
8 | Tool reference documentation |
USER.md |
3.5 | User preferences and context |
Install agent skills from git repositories:
# Via agent tool call
clawhub install research-paper
clawhub install https://github.com/user/my-skill.git
clawhub list
clawhub search "data analysis"Each skill is a directory with a SKILL.md file (YAML frontmatter + markdown body) that gets injected into the system prompt.
See docs/API.md for the complete HTTP, WebSocket, and IPC protocol documentation.
git clone https://github.com/prismer-ai/agent-core.git
cd agent-core
npm install
npm run build # TypeScript compilation
npm test # Run all tests (vitest)
npm run typecheck # Type checking onlyMIT © 2026 Prismer.AI