All selected work
Open-source agent infrastructure

Continuity infrastructure for AI-assisted engineering.

Resumable workflows, persistent memory, and durable project knowledge—designed as one system so complex work can continue without reconstructing its context from scratch.

View public source ↗
3 MCP serversWorkflow · memory · knowledge
3 harnessesClaude Code · Codex · Gemini CLI
9 workflowsBundled reference processes
108 testsCore and server suites

01 · Brief

Complex engineering work rarely fits inside one chat, one agent, or one uninterrupted session. Decisions disappear, partial work loses its operating context, and the next agent is forced to rediscover what the previous one already learned.

Saguaro treats continuity as infrastructure: execution state, durable lessons, and project knowledge remain available across sessions and supported harnesses.

02 · What I owned

Designed the product model and boundaries between workflow, memory, and knowledge; built the shared TypeScript core and three MCP servers; created project-local configuration, cross-harness marketplace artifacts, installation tooling, bundled workflows, documentation, and validation suites.

03 · The system

Workflow

Preserves phase state, artifacts, approval gates, validation results, and the exact workflow definition needed to resume safely.

Memory

Captures lessons and observations at run, project, and global scope, with deliberate promotion between lifecycles.

Knowledge

Stores and retrieves document-scale decisions, specifications, and research through ranked semantic chunks and optional synthesis.

04 · Decisions that mattered

Separate memory from knowledge

Lessons and full documents have different lifecycles, scopes, and retrieval behavior.

Resume from persisted truth

Runs retain phase state, gates, artifacts, and a snapshot of their workflow.

Make boundaries enforceable

Approval gates cannot be cleared by an agent claiming approval; contracts validate required artifacts and context calls.

Redact before persistence

Sensitive material is removed before local storage or embedding.

05 · Outcome

Saguaro is live and used selectively on complex private projects where work spans sessions, agents, and approval boundaries. It preserves the state, lessons, and durable knowledge needed to continue without rebuilding project context from scratch.

The hard part is not making agents more autonomous. It is giving that autonomy durable context and visible boundaries.