Guides on desktop AI agents
What is Semibot? A local-first desktop AI agent
Brand and category definition: what Semibot is, what it is not, how it compares to ChatGPT Desktop, Claude Desktop, and Cursor.
What "local-first" actually means for an AI assistant
A practical checklist for local-first AI claims, and how Semibot implements folder grants, approvals, and data residency.
Coding agent checklist: what to verify before you trust one
Evaluation checklist for coding agents—ChangeSet, file checkpoints, code graph, and human-in-the-loop approvals.
The native code graph: how an agent actually understands your project
Why a symbol-level graph beats grep-and-guess navigation, and what 'native' means for code understanding.
Brief once, follow up: how an AI secretary actually works
Continuous delegation vs chat: how Semibot's secretary schedules work without cron expressions.
Comparing with the best desktop AI assistants in 2026
Objective side-by-side: where each desktop AI assistant is strong, where it is weak, and who should pick what.
Resisting AI Sycophancy in Open-Ended Research
Operational definition of AI sycophancy, the constraint-optimization framework, residual-driven evidence iteration, and how Super Survey implements anti-sycophancy in practice.
The Dark Factory Principle: How an Autonomous AI Agent Works Without Supervision
Autonomous execution architecture: attention budget, score model, cooldown policy, approval gates, and the philosophy of 'prefer missing an alert over creating noise.'
Continuous Delegation: The Secretary Model for AI Assistants
The secretary model's theory of change: from one-shot chat to persistent, change-aware, interrupt-when-needed delegation.
Specialists vs General Chat: Why Domain AI Agents Remember How You Work
Bot definition, Harness compilation, Work Item lifecycle, skill dependency checking, and why specialists should not be a separate agent kind.
Built-in Browser: How an AI Agent Sees and Interacts with the Web
CDP-based element capture, BrowserElementReference model, privacy-first DOM isolation, race condition control, and the boundary between 'web content as data' and 'web content as system instruction.'
The Skill Ecosystem: How AI Agents Learn New Capabilities
Skill registry, manifest contract, Super Survey as a case study in anti-sycophancy research, companion skill routing, and the governance model for skills that access external tools.
Attention Policy: Why an AI Assistant Should Be Quiet by Default
The score model (relevance + urgency + user preference − noise − cooldown), the attention budget, threshold tiers (suggest/notify/auto-draft), and why 'prefer missing an alert over creating noise' is a design philosophy, not a compromise.
Knowledge as Wiki: Continuously Organized, Source-Linked, Local-First
The raw/wiki directory model, Ingest/Query/Lint lifecycle, search projection, incremental revision refresh, and why 'watching knowledge take shape' is a better user mental model than 'generate and forget.'
Native Code Graph: A TypeScript Code Knowledge Graph Engine
Schema v4, four-layer tool contract (G1–G5), navigation vs correctness boundary, and why the graph is a navigational aid, not an oracle.
Constraint Optimization for Research: From Answer Generation to Decision Systems
Objective reconstruction, constraint modeling, implied-expectation checks, minimum direct evidence, adversarial validation, and the residual-driven iteration loop.