Pricing
The local tier is the product. Hosted is the multiplier.
Ganvil's engine and CLI cost nothing, on principle — canonical history should never sit behind a subscription. Ganforge charges for the things that are genuinely org-scale: shared indexes, offloaded compute, and locked-down data flow.
The full product on your machine. Not a trial, not a teaser tier — the free rail is an architectural commitment: everything works offline, with AI switched off entirely.
- ▸gan CLI — 1:1 git drop-in, content-addressed with BLAKE3
- ▸Byte-perfect two-way git bridge (gan import / gan export)
- ▸gan mcp with all 15 agent tools, read-only and deterministic
- ▸Symbol diff, symbol blame, decision archaeology (sdiff / blame --symbol / why)
- ▸On-device embeddings, NPU-first with GPU fallback
- ▸Freshness-stamped answers on every tool response
- ▸Works fully offline · zero AI required
Ganforge hosts the Hammer and the index in the cloud: compute embeddings once and synchronize workspaces remotely. Connect your local machine for an optional fee.
- ▸Everything in Local
- ▸Hosted index: embeddings built once, shared across the org
- ▸Cloud-Hosted Hammer — zero local CPU/RAM pressure for indexing
- ▸Local machine coordination add-on (+$5/seat/mo) — link local hardware/worktrees
- ▸Cross-repo search and cross-repo impact analysis
- ▸Remote MCP gateway with tenant scoping and rate limits
- ▸Silent degradation: if the hosted tier is unreachable, tools fall back to local
Pull mode inverts the data flow: the index travels to your machines and questions stay home. Built for teams whose security review reads network diagrams line by line.
- ▸Everything in Team
- ▸Pull-mode index sync — query text never leaves your machines
- ▸Coordinate unlimited local machines and dedicated infrastructure
- ▸Signature-required proposals: agent writes demand an Ed25519 key from your repo’s trust anchors
- ▸Append-only audit trail of agent activity
- ▸Deployment and migration support from the team that built the engine
None of ours. All of yours — and the product's job is making "yours" small. The data layer spends zero LLM tokens: outlines, callers, impact, symbol diffs are computed from parsing and graph analysis, not from a model. Whatever reasoning happens runs in your model, on your bill, and every table on the benchmarks page is about shrinking that bill.
We never hold LLM keys and we never mark up tokens. There is no margin hiding in your usage, on any tier.
What hosted billing meters: seats, hammer-minutes for indexing offload, and — once Anvil Pods ship — workspace-minutes for per-agent cloud checkouts. The third meter carries the same roadmap label as the pods themselves; it appears here so the meter list is complete, not because you can buy it yet.
A note on the placeholder figures: Team and Enterprise pricing will be set at public launch, and the numbers above exist so the tiers can be compared, not quoted. What each tier buys is fixed; what it costs is not yet. The Local tier's $0 is neither placeholder nor promotional.