Choose theme
AI Resources
Open Instinct
Open Instinct is a personal-agent project built around text messages, connected apps, its own computer, and requests from other people or their agents.
Its code separates what the owner, a partner, family, friends, contacts, and strangers may ask it to do. You can inspect the source and start with local command-line chat; the maintainers describe the live messaging and hosted deployment paths as beta. Use this as a first read, not a recommendation. Open the original project before trusting details like terms, limits, privacy, cost, setup, or safety.
What it is
A personal agent you can text
The project combines a Pi agent runtime with Inkbox messaging, Composio app tools, and optional Maritime hosting and desktop access. A local CLI lets you explore it without provisioning a phone number.
Why it stands out
Requests have different permissions
A friend can ask for free time under the default rules, while calendar details are reserved for the owner and partner. The runtime checks tool calls against these tiers and any active grants.
Availability
Public source, beta integrations
The repository provides Node.js and pnpm setup, service documentation, deployment files, and tests. It declares an MIT license. Expect beta rough edges, including a reported self-hosted messaging-tunnel problem with a fix still proposed.
Why it matters
What makes it useful
A scheduling request can arrive from someone other than the person who owns the assistant. Open Instinct makes that distinction part of the workflow: a friend can ask about availability or propose a plan, but committing a plan asks the owner for approval under the default policy. Builders can examine how a personal assistant handles shared tasks without giving every requester the same tools.
What to know
Where it fits
Start with a Node.js service and a model provider. The documented local path uses instinct init --skip-inkbox, instinct dev, and instinct chat. Inkbox adds the phone and messaging identity; Composio connects app accounts. Maritime is an optional hosting and computer path, and the documentation also allows running without a desktop. These are separate setup choices, rather than one bundled consumer app.
Notable points
What stands out
The assistant's persona, memory, and permission configuration are files you can read. Memory uses Markdown, and tool decisions are recorded in an audit log. The default tier table is also visible in the code: desktop and file tools are owner-only, while an agent acting for a friend receives that person's tier. This gives builders concrete records and rules to examine when understanding a request.
Before using
What to review
The README specifies Node.js 22.19 or newer and pnpm 10. Model access and optional messaging, app, hosting, and wallet services require their own setup and credentials.
An open pull request reports that the self-hosted Inkbox tunnel registers but does not start serving traffic, and that body-bearing webhook forwarding needs a second fix. Check whether that repair has landed before choosing the tunnel path.
The security documentation says the workspace rules are not an operating-system sandbox for the owner's shell, and that prompts from untrusted content can still affect the model. Permission tiers do not establish that an agent's decisions are correct.
The documented hosted setup entrusts data to the model, messaging, app, and hosting providers involved. Read the service and security documentation before connecting accounts.
Reader fit
Who may find it relevant
Most relevant to developers exploring a personal assistant whose contacts and other agents can make requests, and who are comfortable managing a Node.js service and integration credentials. Someone looking for a finished assistant can read the architecture, but should expect installation work and beta behavior rather than a ready-made app.
Editorial note
Why LifeHubber lists it
We list Open Instinct because it makes a common personal-assistant question concrete: what should happen when another person asks your agent for help? Its free-time, calendar-detail, and approval rules show how the same assistant can respond differently to different relationships. That is useful for a builder deciding which shared tasks to allow, because the choice can be examined as tool permissions rather than left only in a conversation prompt.
Source links
Source materials
Reader note
Before relying on this entry
LifeHubber lists entries to help readers inspect AI projects, not to endorse them or prove they are safe, suitable, accurate, maintained, or right for a specific use. We do not verify every entry in depth. Before relying on anything listed, review the original materials, terms, privacy practices, limits, and risks that matter for your situation.
What to explore next
Decide what an assistant may do with your accounts.
Open Instinct exposes tiers for other requesters. Continue with practical checks for the access, approvals, and recovery points around the assistant itself.
More in AI Agents
Keep browsing this category
Explore more AI agent projects.
Paperclip
paperclipai/paperclip
A self-hosted server and dashboard for coordinating agent teams through companies, goals, roles, issues, heartbeats, budgets, approvals, and persistent activity records.
MiroFish
666ghj/MiroFish
A document-based multi-agent simulation project that builds personas, runs simulated social activity, and produces reports and agent conversations. Its application requires a model API and Zep Cloud; generated scenarios do not establish real-world outcomes.
OpenHuman
tinyhumansai/openhuman
A TinyHumans desktop personal agent with a local Memory Tree and editable Markdown vault, visual workflows, agent orchestration, broad app integrations, optional local models, and managed services for sign-in, model routing, search, and some integration paths.
For project maintainers
Listed here? You can use the badge.
If you maintain a project with a current LifeHubber listing, you may add the optional “Listed on LifeHubber AI Resources” badge to its README, docs, or website. No introduction or permission request is needed.
