The Ontology of Context: A Primer

1. Intelligence Without World

A language model drafts a flawless business plan overnight. It cites market data, models cash flow, and names a viable launch date. The team reads, applauds, and then sets it aside. No one knows who would own the next action, which legacy customers would resist, or which undocumented policy would block procurement. The plan is brilliant yet inert because it floats free of the world that must enact it.

Vela begins where that document stalls. Its claim is blunt: value migrates from the ephemeral act to the permanent architecture. The architecture in question is context: recorded, queryable, and continuously re‑applied to every decision.

2. Being‑in‑the‑World: Why Context Is Ontological

In Being and Time Heidegger distinguishes two ways a tool can exist for us.

Most software agents resemble present‑at‑hand implements: they accept instructions, ignore setting, and return detached output. An agent running inside Vela is different. Because it can traverse the institutional graph, check budget limits, and read prior resolutions, it belongs to the workplace that invokes it, it is ready‑to‑hand. Remove that graph and the same agent reverts to a gadget that must be micromanaged.

The shift is ontological, not cosmetic. Context does not decorate action; it makes action possible.

3. Symmathesy: Institutions as Learning Ecologies

Gregory Bateson coined symmathesy to name a system that learns through the relationships among its parts. Learning is not stored inside any single node; it resides in the pattern of interaction itself.

Vela implements symmathesy with a hybrid memory:

• Graph store – an append‑only record of facts, actors, and causal links.
• Vector index – embeddings of conversations, proposals, and outcomes.

Every fresh event flows into both stores, is embedded, and becomes immediately retrievable. Agents query structure and semantics in one call, retrieve precedent, and act with reference to a living ecology rather than a static dataset.

For builders the rule is plain: publish granular events, tag edges with why as well as what, and the ecology will feed itself.

4. The Institutional Graph as First‑Class Entity

Each DAO's event stream is its world‑line: the only authoritative history of what has happened. Governance parameters exert an attractor force: countless micro‑choices converge toward a few permitted macro‑outcomes. Falsifying that graph is costly because every new statement must reconcile with cryptographic provenance and economic budgets. Scarcity arises not from compute but from the difficulty of forging believable context.

5. Encoding Context — Five Pragmatic Patterns

Each pattern anchors an otherwise transient act to a ledger that future agents can inspect.

6. Measuring Context

Context‑to‑Compute Ratio

If prompts swell while context references plateau, debt is accruing.

Divergence Audit
Replay a workflow with its graph connection disabled. Record performance loss. The delta quantifies reliance on embodied history.

A CLI tool—vela‑context‑lint—will automate both checks.

7. From Philosophy to Deployment

1. Bootstrap Create the graph scaffold, mint inaugural credentials, seed one ritual.
2. Governance Checkpoint Add “context debt” to the retrospective agenda; treat it like tech debt.
3. Migration When importing an existing organisation, replay prior events into the graph rather than snapshotting the end state; the line of becoming matters more than the result.

8. The Unasked Question

References