🧠
Ledger
Durable work memory: items, views, code cards, secrets (myco, kuda, arca)
GaiaPut AI agents to work — on your terms
Open infrastructure for agent-driven development
Open infrastructure for agent-driven development
We build open infrastructure so people and agents can own the AI that works for them.
We're building the missing layers between coding agents and the repositories they work on — a ledger (myco, kuda), a WAL (sturn), an orchestrator (peregrin, sepia), and a queue (mimus, cicada). Open, minimal, composable. A stack agents can actually run on.
The principles behind how we build.
The machines. Each a small engine, each owning one mechanism.
Code-map engine for agent workflows: crawls repos into verified per-module cards (repo card, module cards, symbol graph) that agents query instead of re-exploring.
Ephemeral coordination engine for agents: claims, ops, and topics over the git object store, so agents see each other's intent in milliseconds instead of at merge.
Verification engine and critic for repos and agent memory: extracts claims from docs, comments, and cards, verifies them against the code, and repairs the fixable.
CI for your own forge: pluggable push sources queue per-sha pipeline runs on your box — GitHub first, GitLab later — with GitHub Actions workflows as a runner dialect and tamias-native distribution.
Session lifecycle engine for coding agents: one canonical IR and pluggable store drivers to list, load, attach, prompt, and resume sessions across agents (Devin, Claude Code, Cursor, and more).
Event-driven orchestration engine for agent workflows: declarative campaigns of events, nodes, and drivers across your machines, with a sepia-powered session plane on every node.
Records engine for agent work: a myco CLI owns every write to a repo's project/ ledger (todos, issues, proposals, plans, ADRs, handoffs). Markdown is canonical, git history is the audit trail.
Secrets broker for agents: code carries cuora:// references instead of values, the CLI injects secrets as env vars at run time, and every materialization is audited.
Scheduled-work engine: declare calendars in YAML, plan and apply them against live providers with drift detection, ship calendar bundles through tamias, drive them from peregrin campaigns, plus an MCP server so agents can reconcile them.
Workspace gardener for GitHub orgs: clones, syncs, inspects, and generates editor workspace files for every repo of an org under one root.
Package and content manager for the agent factory: fetches, stores, caches, and dispenses content packages (peregrin bundles, myco kinds, kuda domains) via pluggable adapters.
Visualization host for the 9aia tool suite: renders ViewNode IR trees emitted by tamias drivers across CLI, TUI, and web surfaces.
Distributed substrate for agent workloads: a generic entity plane with namespaced ops, snapshots, and locks across your machines. Integrations plug in as entity kinds.
The technologies we reach for, by layer.
We are always looking for passionate individuals to join us. If you're interested in collaborating or contributing to our projects, feel free to reach out!
For inquiries, please contact Luis at [email protected].