May 13, 2026
Where we’re going.
The direction holds; specific dates can move. Every quarter we publish what shipped next to what we still plan, so you can check our word against our work.
What we’re building toward
Three threads we’re pulling on.
Each milestone below is tagged with the thread it advances.
- 01
Private inference
AI runs inside the customer network boundary. No third-party API calls during inference. Data stays where your compliance requires it to stay.
- 02
Cited, audit-ready answers
Every response traces to its source at the chunk level. Every query logs to a tamper-evident audit chain. No black-box answers on high-stakes data.
- 03
Human-controlled export gates
No analysis leaves the boundary without explicit human approval, logged and reviewable. Private by default means it stays inside unless you approve it.
By the quarter
Named milestones.
Each milestone builds on the one before it: private analysis first, then always-on enrichment, then action and proof, then customization.
- Q3 2026 In progress
Private analysis workspace
Bolde. Console / Bolde. AnalystConsole and Analyst launch first. AI runs on dedicated hardware inside your network, with citations on every answer.
- 01 Private inference
- 02 Cited answers
- Q4 2026
Always-on enrichment
Bolde. ScoutsScouts check connected sources for changes. Console stays current between questions.
- 01 Private inference
- Q1 2027
Action and proof
Bolde. Operator / Bolde. AuditorOperator runs repeatable work inside your rules. Auditor keeps a tamper-evident record.
- 03 Export gates
- 02 Cited answers
- H2 2027
Customer-specific model behavior
Bolde. CustomBolde-Small becomes more specific to approved customer workflows and feedback.
- 01 Private inference
- H2 2027
Reversibility
Bolde. RollbaxUndo for approved agent actions, with the audit trail intact.
- 03 Export gates
- 02 Cited answers
Anti-roadmap
Things we won’t ship — not even if asked.
The clearest signal about who we are is what we have decided not to build.
Inference runs on hardware you control. We do not ship prompts to OpenAI, Anthropic, or any cloud model vendor.
See: /faq#what-model-is-under-the-hoodWe do not ship prompts or responses to our infrastructure. That is the point: the work stays inside.
We do not sell into employment-screening pipelines. AI judgment over human careers is a line we will not cross.
No countdown timers we do not honor. No fake limited spots. No urgency we have not earned.
Your data is yours. We will not sell it, share it, or train external models on it. Period.
We do not make money by surveilling our users. The product is the product.
Roadmap updates.
Get Bolde Updates for quarterly roadmap notes and release signals.