About the Product
An AI Agent That Delivers Outcomes
OneWikiMind is a local-first, autonomous AI agent platform built for software teams and solo founders. Submit a high-level goal and it autonomously plans, executes, reviews, and delivers the outcome as real code, documents, or reports — running entirely on your machine.
The Agent Team
Four specialized AI agents collaborate on every mission, each with a distinct role.
Planner
Converts your mission into an executable task graph with confidence scoring and dependency tracking.
Executor
Runs tasks using built-in tools inside an isolated, sandboxed workspace with full context injection.
Reviewer
Gates every task with lint, type-check, test, and security verification before marking it complete.
Coordinator
Manages job progression, agent handoffs, and resume logic across restarts and interruptions.
Mission & Job Control
Full lifecycle management for every goal — from submission to completion.
Submit a mission with optional constraints, deadline, and spending budget
Automatic decomposition into milestones, jobs, and tasks with a dependency graph
Start, pause, resume, cancel, and retry any job at any time
Jobs survive app restarts — resume from the last checkpoint with zero re-briefing
Full real-time timeline and event stream for every running job
Confidence scoring on the task plan before execution begins
Memory & Continuity
The system never forgets. Persistent memory means no re-briefing, ever.
Every conversation and job outcome stored as a structured memory episode
Semantic search over past episodes — relevant context recalled automatically
Nightly consolidation at 2 AM — compresses history, detects contradictions, flags gaps
Context always injected before any agent prompt so the system knows your project
Resume any interrupted project without re-explaining what you were building
Safety & Governance
Four-tier action policy. You always stay in control of what runs on your machine.
Executes automatically — reading files, running tests
Executes with logging — writing files, running permitted commands
Blocked until you approve — destructive file ops, external network calls
Always blocked — secrets access, billing operations, publishing
Path sandboxing — tools cannot escape per-job directories
Command allowlisting — only permitted shell commands run
Every tool call logged with full inputs, outputs, and context
Built-in Tools
Everything the executor needs to work on real software projects out of the box.
Read / write files, list directories, search workspace
Run shell commands (allowlisted for safety) and test runners
Apply patches, commit to Git, view diffs and status
Calculator — decimal-precise math, no code injection risk
Finance data fetcher — market quotes via yfinance
Report generator — markdown rollups of outcomes, cost, and artifacts
Skill Runtime
An extensible plugin system that keeps the executor safe, current, and expandable.
17 built-in skills: file ops, code ops, memory, credentials, approval, artifact writing, and more
Install community skill packages that extend what the executor can do
Every skill is tested in an isolated environment before it can be activated
Auto-rollback if a skill's failure rate exceeds 20% over the last 10 runs
AI Provider Support
Bring your own API keys. No lock-in. Route locally or to the cloud based on task and cost.
Anthropic
Claude 3.5 / 4 family
OpenAI
GPT-4o and o-series
Gemini 1.5 / 2.0 Pro
xAI
Grok 2 / 3
Ollama
Any local model on your hardware
Cost Optimizer: Detects your local hardware (CPU, RAM, GPU/vRAM) and recommends local models when economical. Tracks quality score, latency, and cost per model to continuously improve routing. Full cost breakdown per job.
Notifications
Five notification types. Approval Required is never suppressible.
Type
Opt-out
Job Completed
Available
Job Failed
Available
Approval Required
Never suppressible
Capability Gap Found
Available
Nightly Reconciliation Summary
Available
Platform & Distribution
Native desktop apps and Docker — your choice.
macOS
.dmg installer — native desktop app
Linux
.AppImage and .deb packages
Docker
Single-image deployment for server/team use