High-Signal Context Layer
Specs as a durable, always-available context for agents. Agents pull from your specs instead of rebuilding understanding every session. Hash-first sync keeps token usage low—fewer tokens, lower cost, agents that stay aligned.
Key Features
Specs as Persistent Context
✓
One canonical source
Your specs folder lives in the web app. Agents pull from one place—no scattered copies, no "which version is current?"
✓
No re-explaining in prompts
Agents read specs directly. Document your goals, constraints, and decisions once—agents consume them without you repeating context every session.
✓
Always up to date
Sync keeps web and local in lockstep. Specs evolve; agents always see the latest. No stale context.
✓
Structured, dense context
Specs, plans, tasks—human-readable, agent-parseable. High signal, less noise than raw prompts.
Hash-First Sync
✓
Send hashes first
Phase 1: agent sends hashes only. Server compares; requests content only for paths that actually need updating.
✓
Content only when needed
Phase 2: only changed or new content is transferred. Fewer tokens, less bandwidth, lower cost.
✓
Server-driven protocol
Server decides what to pull and push. Agent doesn't send content the server hasn't requested—no redundant sends.
✓
Delta sync state
syncstate.json tracks path→hash and baseVersionId. Incremental updates; full resync only when needed.
Agent Alignment
✓
Configurable specs path
Name your folder—specs, openspec, docs. Agent config matches your project layout.
✓
Rules from your structure
Connect your agent—rules are generated from your specs structure. Agents know where to read and write.
✓
Continuity across sessions
Specs persist. Agents don't lose context between sessions—they sync and pull from the same source every time.
✓
Single sync endpoint
One sync command—push and pull in one call. Simple integration for Cursor, Claude, Gemini, and more.
Use Cases
Reducing token usage and cost
Specs as persistent context instead of re-pasting in every prompt. Hash-first sync means fewer tokens per session—lower API costs.
Keeping agents aligned without extra work
One source of truth. Agents pull from specs; no need to re-send full context or worry about drift between sessions.
High-signal context over raw prompts
Structured specs—goals, constraints, decisions—are denser than ad-hoc prompts. Agents get clarity without noise.
Multi-session continuity
Specs persist across sessions. Agents sync, read, and continue—no starting from scratch every time.