OpenHarvey
GitHub · OpenHarveyOne contract. One workspace.
Keep the primary agreement, tender documents, attachments, conversations and deliverables together. Read, ask, review and revise with the relevant document context.
A dedicated workspace for contracts & tenders
Understand every contract.
Ground every decision.
OpenHarvey is an open-source AI contract and tender workbench with source-linked review, configurable playbooks, deliverables and sandboxed agents.
Start with a sample document in your own trial workspace. Workspace UI is primarily Chinese.
SERVICE AGREEMENT / 03
Services agreement
Clause 5 · Payment
Upon receipt of a valid invoice and acceptance of services, the customer shall pay the corresponding fees within ninety days.
↗ Agreement · Page 3 · Payment term
Review payment terms against my playbook and create a risk dashboard.
✓ Read source → Apply playbook → Save outputPayment term needs negotiation
The agreement specifies 90 days; the example baseline is 30. Consider negotiating a shorter term and clear acceptance conditions.
↗ View source · Page 3▤ Payment risk dashboard.html↗
SEE IT IN THE WORKSPACE
See the product in action.
Source navigation, flexible deliverables, personal playbooks, extensible Skills and BYOK—connected in one workspace.
01
Check the finding. See the source.
02
Deliver the format your task needs.
03
Your checklist. Your side of the table.
04
Three core Skills. Room for your own.
05
Try it first. Bring your own model next.
BUILT AROUND THE DOCUMENT
Contract work depends on connected context.
Contracts and tenders have clauses, pages, versions and business standards. OpenHarvey connects these relationships across reading, analysis, revision and delivery.
From a finding to its source.
Follow valid citations to the source page or paragraph, with continuous text highlighting. Distinguish the primary document from attachments and check the evidence behind a finding.
Read source and output side by side.
Use a connected document reader, agent conversation and deliverable preview. Ask about selected clauses, verify report evidence and compare suggested revisions.
Deliverables shaped by the task.
Create summaries, risk reports, clause comparisons, HTML dashboards and structured data. Preview or download saved files in supported formats.
Your methods. Your standards.
Skills define methods, references and delivery requirements. Risk playbooks define checks and baselines, with configuration versions retained for execution.
An agent that carries work through.
Built on native OpenCode agents to read documents, use tools and save results. Choose models, extend Skills and inspect tool execution and file receipts.
A PURPOSE-BUILT WORKFLOW
Choose a workspace around the work you do.
If you are comparing general office agents, try these concrete workflows in OpenHarvey. These describe our design, without assuming other tools lack the same capabilities.
| Your task | The OpenHarvey workflow |
|---|---|
| Review a long agreement and attachments | Organize the primary source, attachments and multiple conversations in a contract workspace. |
| Verify a specific finding | Follow a valid citation from the answer or report to highlighted source text. |
| Apply your business standards | Select Skills, risk playbooks and your party perspective for the task. |
| Turn analysis into deliverables | Save a summary, revision comparison, HTML report or structured data; preview it alongside the source. |
SANDBOX / PRIVACY / CONTROL
Room for agents to work.
Clear boundaries around execution.
Cloud mode runs OpenCode in E2B sandboxes, alongside account permissions, document scopes and persistent storage.
Isolated execution
Agent tools and code run in cloud sandboxes. The application checks ownership of workspaces, conversations and files.
Control documents and credentials
Sandboxes receive accessible task materials and rules. Platform model keys stay server-side behind restricted trial credentials. Personal model credentials follow the selected configuration.
Persist results, manage lifecycle
Saved deliverables persist on the server. Sandbox recovery uses checkpoints. Expired trials are revoked and cleaned up according to system rules.
Data path: browser → application and storage → E2B sandbox → selected model service. Sandboxing does not mean offline processing, zero data transfer or compliance certification. Self-hosting the web application does not automatically replace E2B or remote models. Read the data boundaries ↗
POWERED BY OPENCODE
A flexible agent runtime.
A workspace built for contracts.
OpenCode provides native tools, Skills, sessions and execution permissions. OpenHarvey adds document organization, source navigation, review configuration and deliverables.
OpenHarveyDocuments · Citations · Risks · Deliverables
OpenCodeAgents · Skills · Tools · Sessions
E2B / RuntimeIsolation · Checkpoints · Lifecycle
Models & storagePersonal model settings · Persistent results
QUESTIONS, ANSWERED
Before you begin.
What is OpenHarvey?
OpenHarvey is an open-source AI contract and tender workbench with source-linked review, configurable playbooks, deliverables and sandboxed agents.
Which documents and outputs are supported?
Inputs include DOCX, text PDFs, Markdown and TXT. Scanned PDFs require OCR first. Outputs include summaries, risk reports, revision comparisons and custom files in supported formats; previews depend on the format.
Can English-speaking users use it?
The website is available in Chinese and English. The workspace, built-in examples and some report fields are currently primarily Chinese. Full workspace localization and English-task validation are still pending.
Does sandboxing guarantee complete privacy?
No. Sandboxes isolate execution; documents may still be processed by the application server, E2B and the selected model provider. Choose a deployment and providers appropriate to your data.
Is OpenHarvey affiliated with Harvey?
OpenHarvey is independent and is not affiliated with, endorsed by or sponsored by Harvey. It focuses on contract and tender workflows and does not claim feature parity.
Start with your next contract.
Explore an example, then bring your documents and playbook.