Open-source TypeScript runtime for autonomous AI agents — unified graph orchestration, cognitive memory, streaming guardrails, voice pipeline, 21 LLM providers.
npm install @framers/agentos
import { generateText, streamText, generateImage, agent } from '@framers/agentos';
// Text — just set the provider, AgentOS picks the model
const { text } = await generateText({ provider: 'openai', prompt: 'Explain QUIC.' });
// Images — cloud or local (Ollama, SD WebUI)
const poster = await generateImage({
provider: 'stability',
model: 'stable-image-core',
prompt: 'Art deco travel poster for a moon colony',
providerOptions: { stability: { stylePreset: 'illustration' } },
});
// Streaming
for await (const delta of streamText({ provider: 'anthropic', prompt: 'Compare TCP vs UDP.' }).textStream) {
process.stdout.write(delta);
}
// Stateful agent with sessions and personality
const tutor = agent({
provider: 'openai',
instructions: 'You are a networking tutor.',
personality: { openness: 0.9, conscientiousness: 0.8 }, // HEXACO
});
const session = tutor.session('demo');
await session.send('What is QUIC?');
await session.send('How does it compare to HTTP/2?');
Click diagram to expand. 8 subsystems, multi-provider runtime, extension packs.
Text, images, video, music, SFX, embeddings, and speech from one API. Cloud and local backends share the same surface, with fallback chains and provider preferences for load balancing.
mission() API with Tree of Thought planning, multi-source search, grounding verification, and human-in-the-loop review. 3 autonomy modes, 5 provider strategies, and dynamic graph expansion.
Agents forge new tools at runtime — compose (chain existing tools) or sandbox (isolated V8 with allowlists). LLM-as-judge safety review, tiered promotion, portable YAML export.
Full-duplex voice with endpoint detection modes, barge-in handling, diarization, and Twilio/Telnyx/Plivo telephony bridging for production IVR.
Three authoring APIs — AgentGraph, workflow() DSL, mission() — compile to one IR. judgeNode for evaluation, checkpointing for time-travel, streaming events.
8 neuroscience-grounded mechanisms with HEXACO personality modulation. Ebbinghaus decay, spreading activation, Baddeley working memory, GraphRAG retrieval with episodic-to-semantic consolidation.
5-tier pipeline: PII redaction (regex + NLP + NER + LLM), ML classifiers (ONNX BERT), topicality drift, code safety (OWASP), grounding guard (NLI). Sentence-boundary buffered.
Dataset-driven evals with candidates, graders, and experiments. LLM prompt runner and HTTP endpoint runner. Compare baseline vs challenger. Drizzle ORM with SQLite/Postgres.
3-tier semantic discovery: category summaries (150 tokens) → top-5 matches (200 tokens) → full schemas on demand. 89% token reduction. Agents self-discover tools mid-conversation.
Signed event ledger (Ed25519 + SHA-256 hash chain), soft-delete tombstones, revision history, autonomy guard. Merkle anchoring for tamper-evident external verification.
Telegram, Discord, Slack, WhatsApp, Twitter/X, LinkedIn, Bluesky, Mastodon, and custom adapters. Multi-channel routing, social publishing, browser automation, and adapter APIs.
Sealed storage policy, toolset pinning, secret rotation, soft-forget memory. Full provenance audit trail. Deploy agents that cannot be tampered with after initialization.
generateVideo(), analyzeVideo(), detectScenes(), generateMusic(), generateSFX() APIs. 3 video providers (Runway, Replicate, Fal) + 8 audio providers. Fallback chains, scene detection, RAG indexing.
SKILL.md prompt modules for research, developer tools, communication, productivity, security, media, and creative workflows. Semantic discovery finds the right skill per turn.
Bounded self-modification: adapt_personality (HEXACO mutation with per-session budgets), manage_skills, create_workflow, self_evaluate. Ebbinghaus decay ensures unreinforced changes fade.