Orbit AGI.
Make every AI run count.
Orbit AGI is being built to help AI agents make better decisions about how work gets done and how resources are used. Mission Intelligence guides execution. Compute Intelligence makes the effort more efficient.
The system
Two intelligences. One system.
Mission Intelligence determines the approach. Compute Intelligence guides the effort.
Know what the mission needs.
Understand the task, select the right context, choose an execution strategy, and define what proves the work is complete.
- 01Task UnderstandingGauge the mission's scope, complexity, and requirements.
- 02Context IntelligenceIdentify the files, instructions, and information that matter.
- 03Execution StrategyChoose the appropriate approach: Direct, Focused, or Full Mission.
- 04Verification IntelligenceDefine the checks and evidence needed to confirm completion.
Proposed capabilities, in development.
Orbit in action
One mission. Better decisions at every step.
Explore how Orbit AGI is designed to guide AI agents through different levels of complexity.
Illustrative strategy — not live model output.
Orbit AGI
Mission Intelligence
- Task
- Low complexity
- Strategy
- Direct or Focused
- Context
- Affected files, one constraint
- Verification
- Targeted tests, before and after
Compute Intelligence
- Context use
- Skip broad context loading
- Reasoning effort
- Light: one bounded pass
- Allocation
- One stage, no planning overhead
- Reuse
- Keep what's read for the test run
Decision record
- Task Understandingcomplexity
- lowmediumhighOne behavior in one place: the date conversion.
- Context Intelligencecontext
- filesdocsconstraintsThe picker and the date utilities it calls; the stored date format must not change.
- Execution Strategypolicy
- directfocusedfull_missionDirect if the cause sits in one function; Focused if it spans the picker and its utilities.
- Verification Intelligenceverify
- testsbuildevidenceA test for each affected time zone, and the before and after for review.
Mission
Host agent
Does the work with its own foundation model, following the guidance.
Verification
Outcome
A person reviews the evidence. Reviewed results can inform later decisions.
Feeds back into Orbit AGI
Compute Intelligence
More useful work. Less wasted compute.
The same mission, run two ways: where agent workflows commonly waste effort, and where Compute Intelligence steps in.
The whole repository is loaded and then reloaded, model calls run at maximum effort, failed attempts are retried, and everything is checked at the end. A concept, not a measurement.
Reuse what matters.
Keep useful context and completed work available where safe.
Spend effort intentionally.
Match model usage and execution depth to the task.
Verify efficiently.
Apply checks that fit the change without sacrificing correctness.
Compute Intelligence guides how an agent uses its resources. It doesn't change a model's weights or its underlying intelligence.
Integrations
Built around the AI you already use.
Orbit AGI is being developed to complement existing coding agents, bringing mission and resource guidance into the workflows developers already know.
The shared layer
Orbit AGI
- Mission Intelligence
- Compute Intelligence
Meant to deliver mission and resource guidance through each host's supported tools, plugins, and extensions. The host agent and its model still do the work.
Integration targets, not available integrations. Orbit AGI is an independent LocalDesk project: Claude Code, OpenAI Codex, and Cursor are trademarks of their owners, and naming them implies no partnership or endorsement.
How Orbit AGI connects
Local Runtime
Designed to run alongside compatible agents on your computer, keeping mission and resource decisions close to the work.
local-runtime
design: local-first
first target: macOS
inference: local, where technically feasible
integrates with: compatible agent runtimes
would provide: mission and resource guidance
Release
From research to release.
Orbit AGI is moving toward an open-source release focused on practical capabilities, reproducible evaluation, and useful integrations.
- Development
- In developmentResearch and engineering are focused on mission decisions, resource efficiency, and reliable evaluation.
- Open source
- Open-source release plannedPublic code, documentation, and reproducible evaluation results will be shared when ready.
- Integrations
- Integration targetsClaude Code, Codex, Cursor, and LocalDesk are among the environments being considered for supported workflows. None is available yet.
Better intelligence. Real work.
Orbit AGI is being built to guide how AI agents work. LocalDesk brings intelligent workflows, connected tools, and verifiable outcomes into one workspace.
Orbit AGI is in development and is not included in the current LocalDesk release.