How it works

Technical specifications

This page documents the internal systems and behaviors that power blprnt. Each feature is described in technical terms with no marketing hype.

Orchestration Runtime

This is the blprnt secret sauce.

Execution model

Tasks are executed as a state machine with explicit transitions for plan, run, verify, and finalize steps.
Queued work can be paused, resumed, or cancelled without losing the state of ongoing operations.
Tool routing enforces which operations are permitted in each state.

Reliability controls

Timeout handling and step-level error capture provide predictable failure surfaces.
Execution metadata records decisions, tool calls, and outcomes for auditing.

Technical specifications

Orchestration Runtime

Execution model

Reliability controls

Natural-Language to Code

Agent Specializations

Prompt Templating

Project & Session Management

Ask-Question UI

Rapid Code Editing

Middle-Out Heuristics

Multi-Provider LLMs

Memory Store (Project/User)

Secure Vault

Multi-Agent System

Skills System

Project Primer

In-Session Planning

Rapid Code Discovery

Shell Tooling

Reasoning Effort Control

JSON Repair

Sandbox Security

Event Streaming

Orchestration Runtime

Execution model

Reliability controls

Natural-Language to CodeLLM-driven code generation with tool calling and project-aware context.

Natural-Language to Code

Agent SpecializationsPlanner, Executor, Verifier, Researcher, Designer roles.

Agent Specializations

Prompt TemplatingSystem prompt assembly using personality, skills, primer, and directories.

Prompt Templating

Project & Session ManagementCreate/list/rename/delete sessions and projects; history & rewind.

Project & Session Management

Ask-Question UIStructured multi-choice blocking questions.

Ask-Question UI

Rapid Code EditingFile create/read/update/delete + apply-patch diff engine.

Rapid Code Editing

Middle-Out HeuristicsToken pruning by top/middle/bottom zones.

Middle-Out Heuristics

Multi-Provider LLMsOpenAI, Anthropic, Gemini, Grok, mock.

Multi-Provider LLMs

Memory Store (Project/User)Vector-searchable memory with scopes.

Memory Store (Project/User)

Secure VaultStronghold-based key and license storage.

Secure Vault

Multi-Agent SystemSubagents with specialized roles and isolated sessions.

Multi-Agent System

Skills SystemSkill registry, apply/list skills, skill references.

Skills System

Project PrimerProject-level instruction store injected into sessions.

Project Primer

In-Session Planning`todo` tool for step tracking; `plan_*` for persistent plans.

In-Session Planning

Rapid Code Discoveryrg search, directory tree/search, file search.

Rapid Code Discovery

Shell ToolingSandboxed shell execution tool.

Shell Tooling

Reasoning Effort ControlAutomatic “reasoning effort” classifier per turn.

Reasoning Effort Control

JSON RepairRobust JSON repair for tool arguments.

JSON Repair

Sandbox SecurityCapability-based filesystem sandbox.

Sandbox Security

Event StreamingSession & tool events dispatched to UI.

Event Streaming