Trusted context into reviewed agent work.
OpenAnimus is a local-first cockpit for turning repo context and agent runs into evidence, review decisions, GitHub artifacts, and project memory.
First proof
software maintainers
Core loop
evidence -> review -> publish
Boundary
human approval
Task inputs
Risk: medium
Issue #27
Local-first default
Prior decision
Claude Code / Hermes
Runtime
Reviewed work
Routed workflow
Medium risk: use dev + QA, require evidence, then stage approval.
Decision boundary
Agents prepare work and evidence. People approve important outcomes.
The problem
The visible artifact is only part of the work. The rest lives across context, task risk, workflow routing, run history, review comments, and the evidence behind every change.
Project context is scattered across tools, conversations, files, issues, PRs, and prior decisions.
Agents can act, but they often run without task risk, prior decisions, checkpoints, or evidence expectations.
Human review becomes expensive when live state, changed files, wrong decisions and unresolved risks are buried in logs.
How the agent loop works
OpenAnimus gives AI work a concrete loop: trusted context, routed execution, evidence, review, publishing, and memory.
Start from a goal, issue, PR, or build note, then attach the project context the agent should trust.
Choose the workflow, risk level, expected evidence, and execution path before the run starts.
Capture run state, changed files, commands, checks, artifacts, blockers, and unresolved risks.
Turn the run into a review package so a person can approve, request changes, rerun, or block.
Publish approved GitHub work, then record what happened for future runs.
Wedges
OpenAnimus starts where the loop is concrete and reviewable: real repositories, candidate changes, checks, review packages, GitHub artifacts, and project memory.
First proof
OpenAnimus proves the loop on real software work first: goals and repo context become agent runs, candidate diffs, checks, review packages, GitHub artifacts, and project memory.
Direction
The broader personal AI brain direction is long-term. Today, OpenAnimus focuses on proving the reviewed-action loop through software workflows.
Architecture
OpenAnimus integrates pragmatic context sources and execution tools while owning workflow routing, evidence, review, publishing, and memory.
L1
Repo files, issues, PRs, project notes, decisions, and other intentional inputs selected for the task.
L2
Local worktrees and bring-your-own agent tools such as Hermes, Codex, Claude Code, or other runners.
L3
Task risk, workflow routing, run state, changed files, checks, blockers, review packages, and approvals.
L4
Approved GitHub actions, external links, review outcomes, reusable decisions, failed runs, and lessons.
Principles
OpenAnimus is designed for trusted context and reviewed action, so the product boundary matters.
OpenAnimus should work with context you choose to bring in, with clear control over what is captured, stored, shared, or sent to models.
Agents should stage context, changed files, commands, tests, screenshots, risks, and review notes before important work is approved.
OpenAnimus can help prepare and route work, but it should not claim to replace expert judgment or make high-stakes decisions autonomously.
Open source
Open issues, challenge the architecture, and describe the agent workflows where trusted context, evidence, and human review would change what ships.
Built in public, product first
The project is developed openly so the architecture, tradeoffs, and agent workflows can be inspected. The stream is a proof of process, not the product.
Community discussion and workflow feedback.
Short product updates and build notes.
Short answers for the current product direction.
No. OpenAnimus is the review cockpit around coding agents and runtimes. It helps route work, collect evidence, stage decisions, publish approved outcomes, and remember what happened.
That is the long-term direction. The first public proof is narrower: reviewed software workflows for maintainers.
No. OpenAnimus is designed to stage context you choose to bring in, evidence, drafts, and review trails so people can make better decisions.
Software maintainers using AI agents on real project context: issues, decisions, files, PRs, checks, run history, and reviews.