Continue the conversation — chat opens pre-seeded with the current signal, caps, and movement.
Open platform for cloud coding agents: SDK + CLI + Cloud + GUI. Model-agnostic, sandboxed execution, designed to scale from individual use to thousands of parallel agents. Formerly OpenDevin. V1.0 released December 2025 with complete architectural overhaul. Leading open-source autonomous coding agent with 66.4% SWE-Bench Verified (SOTA).
$18.8M Series A (November 2025) led by Madrona. AMD strategic partnership for local AI on Ryzen workstations. 67K+ GitHub stars, 460+ contributors, 4M+ downloads. Engineers from AMD, Apple, Google, Netflix, NVIDIA confirmed as users.
Adoption & Proof Points
- Funding: $18.8M Series A (Nov 2025), ~$24M total
- Traction: 67,200 GitHub stars, 460+ contributors, 4M+ downloads
- Benchmarks: 66.4% SWE-Bench Verified (SOTA), #1 Multi-SWE-Bench, LiveSWEBench leader
- Enterprise engagement: AMD, Apple, Google, Amazon, Netflix, TikTok, NVIDIA, Mastercard engineers confirmed
- AMD partnership: Strategic collaboration for local AI on Ryzen AI PCs
- Customer result: 30x improvement in CVE resolution (days to minutes)
- Self-contribution: ~20% of OpenHands commits now authored by OpenHands
- Pricing:
Recommended Use Cases
- Self-hosted autonomous agents with full infrastructure control
- Large-scale code automation: parallel campaigns across hundreds of repos
- Custom agent development requiring model choice and runtime control
- Teams avoiding vendor lock-in who can accept governance gaps
- Non-regulated workloads where security certification isn't required
Risks & Limitations
- Aug 2025 "Lethal Trifecta" vulnerability enables token exfiltration via prompt injection
- 148-day disclosure-to-acknowledgment timeline demonstrates weak incident response
- No SECURITY.md, no published CVEs, no vulnerability disclosure program
- Enterprise features exist but no third-party validation
- Running safely at scale requires platform engineering (Kubernetes, secrets, network policy, logging)
- Docker dependency adds friction; Windows requires WSL2
- Dual configuration systems cause confusion
- No native semantic code search—relies on LLM context window
- Large codebase comprehension limited vs tools with dedicated indexing
- High token consumption reported (400K+ input tokens for simple tasks)
- Series A in Nov 2025 means limited enterprise support history
- Source-available enterprise features require commercial license
- Enterprise pricing requires sales contact
Capabilities & Integration
Agentic depth: CodeAct 2.1 agent handles multi-step tasks: code modification, shell commands, web browsing, API calls, GitHub issue resolution. 66.4% SWE-Bench Verified with trained critic model and inference-time scaling. Sub-agent delegation enables hierarchical coordination. Security Analyzer provides LLM-based risk assessment before execution.
Context handling: LLMSummarizingCondenser achieves 50% cost reduction via context summarization. Memory persistence enables save/restore for long-running workflows. Event logs record all actions. Gap: No semantic code search—context relies on LLM window, not embeddings.
Extensibility: Full MCP support (SSE, HTTP, stdio) enables integration with external tools/APIs. Model-agnostic via LiteLLM (Claude, GPT, Gemini, DeepSeek, local models). Skills system supports .cursorrules and agents.md formats. Custom tools follow action/observation patterns.
Integration: GitHub, GitLab, Bitbucket, Azure DevOps. CI/CD via GitHub Actions. IDE support (experimental): VS Code, JetBrains, Zed. Jira, Slack, Linear integrations. OpenHands LM 32B runs on single 3090 GPU.