1. Mike — open-source legal AI
Built in two weeks by a former Latham & Watkins associate. Claims feature parity with Harvey (valued at $11B) and Legora (valued at $5.5B). Document review, contract drafting, citations, tabular extraction, multi-step workflows. Self-host it inside your law firm so files never leave your perimeter.
The pitch is brutal in its simplicity. Why are firms paying enterprise license fees for closed-source legal AI when one engineer can replicate the entire web app in a fortnight and give it away?
It is not going to kill Harvey overnight. But it absolutely changes the negotiation.
2. OpenCode — the AI coding agent that lives in your terminal
github.com/opencode-ai/opencode
150K+ stars. 850 contributors. 6.5M monthly developers. Open-source AI coding agent that runs in your terminal, IDE, or desktop. Works with Claude, GPT, Gemini, local models via Ollama, anything you want.
This is the open-source answer to Cursor and Claude Code. Plan mode for safe exploration. Build mode for execution. LSP integration. Multiple sessions. Vim-like editor. Total flexibility on which model you use, which means zero vendor lock-in as the model layer keeps commoditizing.
If you are paying $20/month for Cursor and you have not even tried OpenCode, you are leaving money and optionality on the table.
3. Hermes Agent — the agent that grows with you
github.com/NousResearch/hermes-agent
Built by Nous Research. The only open-source agent with a built-in learning loop. It creates skills from experience. Improves them during use. Searches its own past conversations. Builds a deepening model of who you are across sessions.
It is not tied to your laptop. Run it on a $5 VPS or a GPU cluster. Talk to it from Telegram, Discord, Slack, WhatsApp, Signal, email — 18 platforms out of the box. Use any model. Switch with one command.
The Curator release is what makes this special. Hermes maintains itself. A background agent grades your skill library on a 7-day cycle, consolidates related skills, prunes dead ones. The agent gets better the longer you run it.
This is what most VC-backed agent startups are still trying to build. It is open source.
4. Multica — managed agents platform
Open-source infrastructure for treating AI coding agents as actual teammates. Assign them issues like you would a colleague. They pick up the work, write the code, report blockers, update statuses, ship.
Works with Claude Code, Codex, GitHub Copilot CLI, OpenCode, Hermes, Gemini, Cursor Agent, Kimi, Kiro CLI. Vendor-neutral by design. Self-hosted by default. Code never passes through their servers.
The bet is multiplexing. For decades, software teams have been single-threaded. One engineer, one task, one context switch at a time. Multica says a small team should not feel small. Two engineers and a fleet of agents should move like twenty.
This is the operational layer that most companies are duct-taping together with Linear plus a Slack channel plus copy-pasted prompts. Multica replaces the duct tape.
5. Promptfoo — AI security and red-teaming
github.com/promptfoo/promptfoo
156 of the Fortune 500 use it. 300,000+ developers. Just got acquired by OpenAI.
Promptfoo simulates real users to find application-specific vulnerabilities in your AI apps. Direct and indirect prompt injections. Jailbreaks tailored to your guardrails. Data and PII leaks. Business rule violations. Insecure tool use in agents. Toxic content generation.
If you are shipping anything with an LLM in it and you are not running Promptfoo against it before deploy, you are shipping vulnerabilities by default. Anthropic uses it. OpenAI bought it. That should tell you everything.
The pattern
Every one of these tools either replicates a billion-dollar incumbent or builds infrastructure most VC-backed startups are still pitching as their core product. Open source. Free. Self-hostable. Vendor-neutral.
Models are commoditizing. The premium on closed-source AI tools is collapsing in real time. The companies that compound from here are the ones building on infrastructure they own and contributing back to the layer everyone runs on.
If you are still paying for the closed-source version of any of these, you should at least know the open-source alternative exists.
Star the repos. Fork the ones you need. Tell your team.
Building the intelligence layer underneath your AI stack? That's what we're working on at ELI. Payroll for your AI workforce.