Questions, answered like a human
Things you might want to know.
No futuristic word salad. Just what Willow does, how the memory works, what it costs, and what is still being built.
Willow Basics
Willow keeps long conversations with AI on track, it holds on to what you're trying to do, what you've ruled out, and what's already been decided, so the thread you started is still the thread you're in.
No. Willow isn't a chatbot or a foundation model. It works alongside the models you already use, keeping the thread steady around them. The model writes the words; Willow keeps the intention.
No. Memory tools store what was said. Willow carries forward what matters, the goal, the limits, the decisions, the open questions, in a compact form, rather than replaying whole conversations back into the prompt.
Anyone whose thinking happens over more than one sitting: founders, researchers, students, parents planning something complicated, professionals with long projects, and people who find it draining to re-explain themselves. Developers can build on the same engine directly.
AI systems lose the thread across long conversations, tool workflows, model switches, and fresh sessions. That means repeated prompts, bloated context, stalled agents, inconsistent answers, and manual repair. Willow is designed to reduce that.
How It Works
Willow keeps a short, living summary of what the conversation is actually about, the goal, the constraints, the decisions made, the questions still open, and carries that forward instead of resending the whole transcript.
Willow compares each new answer against what the thread was meant to be doing, and forms a view on whether it's still on track. If it isn't, you get a clear next step rather than a silent wrong answer.
No. In the Observatory it happens quietly while you work. Developers who want the underlying signals can read them directly through the gateway.
No. Willow is built around sending the relevant thread state rather than replaying full chat history.
No. Willow improves how a thread is held and carried forward, but it doesn't replace clear instructions, good judgement, or checking important answers.
Access
Willow Observatory is the public application, create an account and start a thread. Builders can reach the same hosted continuity layer directly through the gateway, MCP server, and Python SDK, on Root, Canopy, or Forest.
Yes, you can sign up for Willow Observatory directly and begin using it right away.
Yes. At launch you need to be at least 18 to create or use a Willow Observatory account, and you'll be asked to confirm that during signup. Your birth month and year are checked in your browser only, then discarded, we don't receive or store them. All we keep is a record that the 18+ threshold was met, which policy version applied, and when. There's no under-18 or guardian account today; we may look at that later, but it isn't something we're promising.
Three plans, all on the pricing page: Willow Root ($28) is a one-month builder pass into the hosted layer, Willow Canopy ($98/month) is ongoing developer access, and Willow Forest is for teams, priced on request. Usage is metered in Investigate credits. Current plan details are also shown in-app when you create an account.
Willow currently offers API/proxy access, MCP support, and a Python SDK. Additional SDKs, connectors, and platform integrations may follow based on developer demand.
Yes. Willow is designed to sit in front of or beside existing AI workflows. Developers can integrate through API/proxy patterns, MCP tools, or SDK calls depending on their architecture.
MCP, SDK, and API
The MCP server exposes Willow continuity tools to compatible AI development environments. It allows tools and agents to stabilize prompts, export/import continuity artifacts, run continuity checks, and access coherence signals.
The SDK gives developers a simple way to call Willow from application code, inspect coherence fields, and build continuity-aware workflows without manually handling every API request.
The API/proxy lets developers route AI requests through Willow while preserving compatibility with common model-call patterns. It is useful for teams that want continuity signals and context steering without rebuilding their whole application.
Use the API/proxy if you want model-call routing. Use MCP if you want Willow available as a tool layer inside compatible agent environments. Use the SDK if you want direct application-level integration. Many teams may use more than one.
Metrics and Evidence
Current controlled private tests include 85.4% compact-artifact token reduction in a long-context compression test, 2.25x fresh-session continuity resume lift, reduced measured drift in provider tests, and live API/MCP/SDK smoke-test validation.
Not yet. These are controlled private tests and early validation results. Public benchmark packaging, methodology documentation, and broader evaluation are in progress.
No. Willow can reduce repeated context and manual repair in workflows where context bloat and fragmentation are real costs, but actual savings depend on workload, model pricing, integration design, and usage patterns.
No. Willow does not eliminate hallucinations. It is designed to improve continuity, context alignment, and coherence visibility. Developers should still evaluate model outputs, use verification where needed, and apply domain-specific safety controls.
No. Willow does not claim zero drift. Willow exposes drift/coherence signals and is designed to reduce fragmentation risk, but model behavior still depends on provider, prompt, task, and context.
Security and Privacy
What's stored depends on how Willow is deployed and integrated. Thread state and usage records may be kept to run the service, support you, and keep it reliable. See the Privacy page for the fuller picture, and write to support@willow-intelligence.com to request a copy or deletion.
No. Provider keys stay server-side; Willow is designed so credentials are never exposed in frontend code.
Email support@willow-intelligence.com with enough detail to reproduce it. We'll confirm receipt before investigating. The Security & Responsible Use page describes what we ask of reporters.
Private, team, and organizational deployment options are being evaluated. Teams with specific security, compliance, or infrastructure needs can reach us through the support page.
Getting Help
Use the Contact & Support page, or email support@willow-intelligence.com directly. A person reads every message.
We aim to answer within a couple of business days, and sooner when someone can't get into their account. We don't publish a guaranteed response time we can't hold ourselves to.
The email address on your Willow account, roughly when it happened, and what you expected instead. Never send passwords, API keys, or card numbers.
Roadmap
Near-term priorities include stronger developer documentation, expanded evidence reporting, improved SDK examples, MCP packaging, and broader integration testing.
No. Willow is the continuity layer, hosted, nothing to download. Willow Observatory is the application where people experience that continuity, talk, investigate, organize, build, and return to the thread.
Investigate is a capability inside Willow Observatory. When a question needs more depth, it brings in additional model perspectives, keeps your goal and constraints intact, shows where they disagree, and returns one coherent answer, without leaving the Observatory.
Start in Willow Observatory with a real situation where your AI usually loses the thread, and compare what happens with what happens today.
Still stuck?
Ask us directly.
If your question isn't here, a person will answer it. Or start in the Observatory and see how the thread holds for yourself.