Your AI agents
share no state
by default.
Spelunk is the cross-harness memory layer for AI coding agents. Open-source, code-aware, local-first. Cloud adds real-time coordination across every agent on your team.
From index to coordination
in three steps.
The Spelunk CLI parses your codebase with tree-sitter into a code graph — callers, callees, imports, structure. Hybrid BM25 + semantic search runs locally. No cloud calls. No API keys.
tree-sitter AST · code graph · hybrid search
Decisions, intents, and conventions land in
spelunk memory — stored in git notes so memory
travels with the repo. Every entry has
commit-SHA provenance. Agents call
spelunk memory at session start and
inherit the full context.
A single engineer's agents already share context. That breaks the moment work crosses machines. Spelunk Cloud streams memory updates across every engineer, every agent, in real time — so when one agent learns something, the rest of the team sees it within a heartbeat.
live SSE stream · cross-engineer · cross-harness
Built differently,
by design.
Every architectural decision in Spelunk exists to solve a specific problem agent teams hit at scale. Nothing is bolted on.
Tree-sitter AST chunking at function and class boundaries across 20+ languages. Plus a code graph for callers, callees, and imports. Agents reason about structure, not just tokens.
The OSS CLI runs entirely on your machine — code never leaves it. The OSS Server is self-hosted behind your firewall. Cloud is opt-in for teams that want managed deployment and live cross-agent coordination.
Decisions, intents, and architectural choices persist across sessions. New entries supersede old ones rather than overwriting — so the history of why something changed is always recoverable. Every entry has commit-SHA provenance.
Speaks NDJSON over stdout, exit codes 0/1/2, optional REST. Works with Claude Code, Cursor, Continue, Copilot, or anything that can shell out. Your memory layer doesn't change when your agent does.
Built for code.
Not just conversations.
Four decisions made early that shape how Spelunk works.
- Code-aware over chat-aware. Code has shape — functions, call graphs, dependency chains, git history. We index that shape, not just the words.
- Local by default. The CLI works fully offline. The OSS Server runs behind your firewall. Cloud sync happens only when you opt in.
- Memory that compounds. Every architectural decision captured makes the next session faster. Your team's context gets richer over time — and travels with the repo via git notes.
- Open at the core. The CLI and the Server are MIT-licensed. You can read the code, run them yourself, and never have to bet on a black box.
Start with the CLI.
Grow with your team.
Hybrid base + included operations + soft-cap overage. We meter repos and operations — not humans, not agents. Bring your own LLM keys.
The CLI and the self-hosted Server. MIT licensed. Local code intelligence, AST indexing, hybrid search, memory — zero cloud dependency.
- Local AST indexing & code graph
- Hybrid BM25 + semantic search
- Memory with supersede & harvest
- Self-hosted Server (Docker)
- Community support
Managed deployment with real-time agent coordination. For small teams and consultancies on a single project.
- Everything in OSS
- 5 users · 3 repos · 30k ops / mo
- Live SSE memory stream
- Hash-chained audit log
- SSO / identity
- EU hosting (GCP eu-west)
Mid-size teams running multiple repos. Soft-cap overage at £8 per additional 10k ops — no surprise bills.
- Everything in Starter
- 15 users · 8 repos · 120k ops / mo
- £8 / 10k ops overage
- Priority support
- Enterprise tier from £2,500 / mo
Be first when
we open the gates.
Spelunk Cloud is in private beta. UK and EU teams first. Drop your email and we'll reach out when access opens for your team size — one email, no list, no spam.