Conversation

Language is one organ of the persistent system. Messages, traces, sources, outcomes, and later effects remain separate and inspectable.

Cognitive Field

A view of current recurrent activity: causal traces, typed unresolved questions, and thought packets. Ordinary user questions are no longer treated as autonomous research tasks.

Recent cognitive traces

Open questions

Thought packets

Autonomous Work

Conversation is an interruptible activity, not the scheduler. Work may be completed without constituting development; concrete progress is tracked separately when a hypothesis, experiment, organ, compiler, or validated artifact actually changes what SeedMind can do next.

Work queue

Self-audits

Development Lab

The closure layer. A development chain is expected to move through measured issue → falsifiable hypothesis → experiment → evidence. Reading and auditing alone are not capability gains.

Development objectives

Measured issues

Hypotheses

Experiments & candidate artifacts

Concrete progress lineage

Memory & Goals

Goal ownership is part of the representation, not a label the language model may freely invent. Owner goals require direct owner evidence; SeedMind goals and autonomous work remain explicitly separate.

Your goals

Shared commitments

SeedMind goals

Unverified / legacy goal claims

Memory ecology

Unresolved questions

Research & Learning

Web search is reserved for externally checkable factual uncertainty. Questions about you, communication preferences, or SeedMind's own behavior stay inside interaction and reflection rather than being outsourced to the web.

Manual research

Research sessions

Candidate beliefs

Crystallization candidates

Organs

Built-in organs are infrastructure. Crystallized micro-organs are tiny programs SeedMind may design itself, validate against a narrow safe ABI, run in shadow, and only later promote after observed competence.

Built-in organs

Generated micro-organs

Recovery

Candidate saving, structural validation, functional attestation, golden promotion, and restoration are deliberately separate powers.

Snapshot lineage

Compute & Resources

SeedMind treats compute as a finite cognitive resource. Language inference prefers the selected physical accelerator when a verified accelerator runtime is available; deterministic organs, storage, web work, and compilation remain on CPU. Under pressure, autonomous work normally yields rather than monopolizing the machine.

Detected accelerators

Resource policy

Allow full autonomous compute

Bypass CPU/GPU utilization ceilings for background tasks. Battery protection, owner-interaction priority, model-call budgets, self-modification limits, and recovery protections stay active.

Setup

SeedMind can install its local inference runtime and a selected GGUF language model directly. The host and memories remain usable even when the language organ is absent.

Model catalog

Installed GGUF models

Download and installation jobs

Large downloads: model files are approximately 1–2.5 GB. Downloads are placed in models, and official SHA-256 values are checked before selection.

Remote Access

The direct SeedMind listeners remain on localhost. Remote traffic enters only through the authenticated proxy listener, never by opening the normal local port to the internet.

SeedMind login

Configure these credentials locally. They are required on every proxied/remote connection; direct access from this computer remains frictionless. Failed remote logins are rate-limited.

Private phone access

Tailscale Serve exposes SeedMind only inside your tailnet. Install/sign into Tailscale on the computer and phone, then use the HTTPS address.

Anywhere with only the SeedMind login

Tailscale Funnel publishes the authenticated proxy through HTTPS to the public internet. Your phone does not need to join the tailnet, but anyone can reach the login page, so use a strong unique password.

Diagnostics

System state

Log
No log loaded.