Why We Open Sourced 90% of Our Agent Runtime
The reasoning behind giving away the runtime and charging for the platform.
We open sourced the core of Amodal: the reasoning engine, the tool system, the skill loader, the knowledge base, context management, and the CLI. The thing that actually makes your agent work. For free. Forever.
This is not a marketing stunt. Here's why we did it.
The runtime is not the moat
Most agent frameworks follow a similar core logic: take a user message, load some context, call a model, parse the response, maybe call some tools, loop until done. The ReAct loop is well-understood. The implementation details matter (context management, loop detection, smart compaction), but they aren't defensible as proprietary IP.
This led us to open source the runtime and concentrate on what truly compounds.
The runtime, CLI, and dev experience are free. The platform is where we make money.
What actually compounds
Every connection, skill, and automation published to the registry makes the platform more valuable. A Salesforce package used by 200 companies is better than one used by 1, because 200 companies have filed bugs and validated coverage. This is the Terraform provider dynamic: 20 at launch became 3,000+.
Every session can propose knowledge updates. After 50 sessions, the KB has learned the customer's real patterns, baselines, and edge cases. Session 51 is materially faster than session 1. This compounding knowledge is tenant-specific and lives in the platform.
Observability, session replay, eval dashboards, the Model Arena, team management, SSO, audit logging. None of this is in the open-source runtime. All of it is what enterprises pay for.
Ecosystem growth: the Terraform precedent
Terraform providers over time (community + first-party)
The Terraform analogy
HashiCorp open sourced Terraform. The CLI is free. The state management, collaboration features, policy engine, and registry are the paid product. HashiCorp went public on this model. Our split is the same:
A developer can clone the repo, run amodal dev, and have a working agent on their machine in 5 minutes. No account. No credit card. No trial period. The agent works, locally, forever.
When they want to deploy it to their team, share it on Slack, track costs, run evals, manage access, and have the knowledge base learn from every session — that's the platform.
Trust as a growth strategy
Developers don't trust closed-source AI infrastructure. They've been burned too many times. The vendor changes pricing. The vendor gets acquired. The vendor deprecates the API you built on.
Open source is how you earn trust with technical buyers.
The open source funnel
Developer discovers Amodal (HN, Twitter, conference, word of mouth)
Runs npx amodal init — working agent in 5 minutes. Free.
Shows it to their team. Team wants Slack, cost tracking, access control.
Runs amodal deploy. On the platform. $49/agent/month.
Agent gets better with use (knowledge flywheel). Switching cost increases organically.
What we keep closed
Base prompt templates, compaction algorithms, loop detection heuristics. These deploy server-side without SDK releases. They get better continuously and every customer benefits automatically.
Session replay, eval dashboards, Model Arena, team management, SSO, audit logging. Collaboration and observability features that teams need and individuals don't.
The registry, package hosting, version management, verified badges, install counts. The ecosystem that makes the packages valuable.
The bet
We're betting that a thriving ecosystem is worth more than a proprietary runtime.
More packages, more contributors, more trust, faster adoption. That's the value of open source. The alternative (higher per-unit margin, stronger IP moat, more control) sounds good on paper but doesn't compound.
Terraform, Docker, Supabase, Vercel, and Next.js all made the same bet. It worked. The agent runtime space is moving too fast for proprietary moats to hold. The ecosystem moat is the one that compounds.
The Amodal runtime is open source under Apache 2.0.