Build production AI agents from any document set.
Contract Address : 0x49e6199efff8af71496637302c5caaa150c84444
- Overview
- Quickstart
- Repository Structure
- Contributing
- Documentation
- Nightly Builds
- Security and Bug Reports
- Contact
- Supported Wallets, Protocols, and Frameworks
- License
- Legal and Privacy
Agentora is a General Doc-to-Agent Studio that turns raw documents into deployable AI agents.
Upload research papers, books, product docs, SOPs, transcripts, and notes. Agentora designs the agent, configures retrieval, wires tools, runs evals, and prepares deployment targets.
- Multi-format ingestion (
pdf,docx,md,txt,csv,html,json) - Automatic chunking and metadata extraction
- Hybrid retrieval (dense + keyword + rerank)
- Source-cited answers with page/section grounding
- Agent templates (
Tutor,Analyst,Support,Research,Ops,Compliance) - Prompt compiler (persona, constraints, and output contract)
- Safety policies and action allowlists
- Built-in evaluation packs (faithfulness, relevance, action safety)
- Versioned agent releases (
v1,v2, rollback) - Multi-channel deployment (Web widget, API, Discord, Telegram, Slack)
- Usage analytics and quality telemetry
- Human-in-the-loop approval workflows
- Wallet lifecycle operations (
create_wallet,import_wallet,list_wallets) - Onchain execution controls (
transfer_asset,swap_asset,deploy_contract) - Policy-enforced action routing (
policy_guard,approval_gate) - Full-trace observability (
export_trace,audit_logger, metrics pipeline) - Workflow scheduling and run recovery (
schedule_action,run_recovery)
| Tool | Purpose |
|---|---|
file_loader |
Load and normalize source documents |
ocr_extractor |
Extract text from scanned PDFs/images |
schema_mapper |
Convert raw docs into structured records |
chunk_planner |
Build chunk strategy per file type |
metadata_tagger |
Add source, date, owner, and security labels |
embed_indexer |
Build and update vector indexes |
keyword_indexer |
Build lexical search index |
retrieval_router |
Route question to best retriever strategy |
citation_resolver |
Attach verifiable references |
fact_checker |
Validate claims against source corpus |
prompt_compiler |
Build system/developer/user prompt layers |
tool_registry |
Register and validate callable tools |
policy_guard |
Enforce safety and data policies |
eval_runner |
Run scenario-based and regression evals |
agent_deployer |
Publish to API/widget/channel targets |
monitoring_sink |
Capture traces, feedback, and errors |
cost_optimizer |
Route workloads by model cost/latency |
cache_manager |
Semantic and response caching |
workflow_orchestrator |
Multi-step agent execution graph |
approval_gate |
Human approval before sensitive actions |
wallet_manager |
Create, import, and manage agent wallets |
onchain_executor |
Execute onchain actions with policy checks |
event_listener |
Watch onchain and offchain events |
workflow_scheduler |
Schedule recurring autonomous tasks |
secret_vault |
Secure API keys and signing material |
tenancy_guard |
Enforce per-tenant resource isolation |
audit_logger |
Persist immutable action and trace logs |
billing_meter |
Track usage per workspace and agent |
webhook_dispatcher |
Trigger external systems from agent events |
run_recovery |
Retry and recover failed workflow runs |
Node.js SDK:
import {
createAgentFromDocs,
updateAgent,
deleteAgent,
listAgents,
ingestFiles,
syncDataSource,
buildIndex,
rebuildIndex,
searchKnowledge,
askAgent,
streamAgent,
runEval,
compareAgentVersions,
deployAgent,
getAgentMetrics,
createWebhook,
rotateApiKey,
createWallet,
importWallet,
listWallets,
transferAsset,
swapAsset,
deployContract,
callContract,
scheduleAction,
createWorkflow,
runWorkflow,
listWorkflowRuns,
pauseWorkflow,
resumeWorkflow,
ingestUrl,
ingestNotion,
ingestConfluence,
upsertDocuments,
createKnowledgeBase,
listKnowledgeBases,
exportTrace,
} from "agentora";Python SDK:
from agentora import (
create_agent_from_docs,
update_agent,
delete_agent,
list_agents,
ingest_files,
sync_data_source,
build_index,
rebuild_index,
search_knowledge,
ask_agent,
stream_agent,
run_eval,
compare_agent_versions,
deploy_agent,
get_agent_metrics,
create_webhook,
rotate_api_key,
create_wallet,
import_wallet,
list_wallets,
transfer_asset,
swap_asset,
deploy_contract,
call_contract,
schedule_action,
create_workflow,
run_workflow,
list_workflow_runs,
pause_workflow,
resume_workflow,
ingest_url,
ingest_notion,
ingest_confluence,
upsert_documents,
create_knowledge_base,
list_knowledge_bases,
export_trace,
)
Agentora ships with a full SDK + CLI workflow for doc ingestion, indexing, evaluation, and deployment.
Install:
npm install agentoraCreate an agent:
import { createAgentFromDocs, deployAgent, runEval } from "agentora";
const agent = await createAgentFromDocs({
name: "Research Copilot",
template: "analyst",
files: ["./docs/whitepaper.pdf", "./docs/tokenomics.md"],
citations: true,
safety: {
piiRedaction: true,
actionPolicy: "allowlisted_tools_only",
},
tools: ["web_search", "calculator", "sheet_reader"],
});
await runEval({
agentId: agent.id,
suite: "default-grounded-v1",
});
await deployAgent({
agentId: agent.id,
target: "web_widget",
});Agentora CLI:
agentora init
agentora ingest ./docs --agent research-copilot
agentora build --agent research-copilot
agentora eval --agent research-copilot --suite grounded-v1
agentora deploy --agent research-copilot --target web_widget
agentora monitor --agent research-copilotInstall:
pip install agentoraCreate and query an agent:
from agentora import create_agent_from_docs, ask_agent, run_eval
agent = create_agent_from_docs(
name="Support Assistant",
template="support",
files=["./docs/help-center.pdf", "./docs/faq.md"],
citations=True,
tools=["ticket_lookup", "status_page"],
)
result = ask_agent(
agent_id=agent.id,
message="How do I reset billing permissions for enterprise users?",
)
print(result.answer)
print(result.citations)
run_eval(agent_id=agent.id, suite="support-regression-v1")Repository layout:
.
+-- examples/ # Example inputs and notebooks
+-- rebiber/ # Python engine package and data assets
+-- scripts/ # Maintenance and data scripts
+-- README.md
+-- pyproject.toml
+-- LICENSE
Agentora architecture:
agentora/
+-- api/ # REST/WebSocket APIs
+-- cli/ # CLI entry points and commands
+-- ingestion/ # Readers, OCR, parsers, sanitizers
+-- indexing/ # Embeddings, vector DB, lexical index
+-- retrieval/ # Query planning and retrieval routing
+-- orchestration/ # Agent graph and execution policies
+-- templates/ # Agent blueprints by use case
+-- tools/ # Built-in and custom tool adapters
+-- eval/ # Offline/online eval suites and scoring
+-- deploy/ # Widget/API/channel deployment
+-- observability/ # Tracing, logging, and analytics
+-- auth/ # API keys, RBAC, and tenancy
+-- billing/ # Usage metering and subscription hooks
+-- sdk/
| +-- js/
| +-- python/
+-- docs/
Contributions are welcome across platform, SDK, and agent templates.
- Open an issue for feature proposals or architectural RFCs
- Submit focused PRs with tests and clear implementation notes
- Keep public APIs and CLI behavior consistent across updates
- Include benchmark/eval deltas for retrieval or generation changes
Recommended local workflow:
pip install -e .[dev]
pytest
ruff check .Documentation set:
- Product overview: this README
- Platform guide: end-to-end build and deployment workflow
- CLI reference:
docs/cli.md - SDK reference:
docs/sdk/ - Architecture notes:
docs/architecture/ - Eval cookbook:
docs/eval/
Nightly builds are published for rapid iteration on new features.
Nightly channels:
pypi:agentora-nightlynpm:agentora@nightly
Stable and nightly tracks are both available.
Security is critical for document-grounded agents and action-enabled tools.
- Report security issues privately before public disclosure when possible
- Open public issues for non-sensitive bugs with clear reproduction
- Include affected module, inputs, expected behavior, and logs
Security focus areas:
- Prompt injection resistance
- Data exfiltration prevention
- Tool call authorization
- Tenant isolation
- Secret handling and rotation
- GitHub Issues: Report bugs or request features
- Maintainer contact: yuchen.lin@usc.edu
The Doc-to-Agent core is domain-agnostic and supports both general and crypto-focused templates.
| Category | Support |
|---|---|
| Wallets | EVM-compatible wallets via adapter plugins |
| Protocols | HTTP, WebSocket, webhook callbacks, function/tool calling |
| Frameworks | LangChain, LangGraph, custom Python pipelines |
| Models | OpenAI-compatible APIs, local model runtimes |
| Vector Stores | FAISS, Chroma, Pinecone, Weaviate adapters |
| Deploy Targets | Web chat widget, REST API, Slack, Discord, Telegram |
| Data Sources | Local files, cloud storage, Notion/Confluence adapters |
This project is licensed under the MIT License. See LICENSE.
Agentora is designed for lawful use with documents you are authorized to process.
- Respect copyright, privacy, and data-processing regulations
- Obtain consent before uploading sensitive user or enterprise data
- Review all generated outputs before production use
- Add human approval gates for high-risk actions