v0.5 public alpha · AGPL-3.0

Design proteins
in your terminal.

fova is a terminal agent for de novo protein design. It plans, runs, and ranks design jobs, then ships the survivors to a wet lab, all from a single Go binary.

Download fova View on GitHub

▸ macOS · Linux · one static binary, no runtime

fova · protein design agent session

The fova promise

Free by default.
No account needed.

Every feature works without a paid account. Local LLMs like Ollama, vLLM, and LM Studio run out of the box, and the whole free knowledge stack needs no keys. Paid LLMs and wet lab submission are the only opt-ins, and they stay optional.

Local LLMs Literature retrieval Design & scoring SQLite provenance Paid LLMs · opt-in Adaptyv wet lab · opt-in

The loop

From a sentence to the bench.

Describe a target in plain language. fova does the rest, and tracks every design back to the intent that produced it.

⌖
01 ▸ STEP

Plans

Turns a natural-language target into a grounded DesignPlan, backed by current literature from Europe PMC, OpenAlex, Semantic Scholar & bioRxiv.

⠿
02 ▸ STEP

Orchestrates

Runs experimentally validated tools like BindCraft, RFdiffusion, ProteinMPNN, and AlphaFold3, locally or on your own Modal GPUs.

≋
03 ▸ STEP

Ranks

Scores every design on interface and confidence metrics, then filters to a small, trustworthy shortlist.

✓
04 ▸ STEP

Ships

Sends the survivors to the bench through the Adaptyv Foundry API, then closes the loop with real results.

Why fova

Principles, not features.

fova is built on a handful of principles. These six are the ones you feel every time you run it.

⊹

Free by default

Local LLMs work out of the box. No account, no key, no card. Paid LLMs and wet-lab submission are opt-in.

✓

Validated tools only

Every built-in design tool has documented wet lab success. No unproven methods ship in the box.

≋

Ranked on real metrics

fova scores every design on multiple interface and confidence metrics, never a single number, so the shortlist holds up at the bench.

⌗

Provenance everywhere

Every design carries its full lineage in local SQLite: intent → tool versions → wet lab result.

◇

Local-first, cloud-elastic

Runs offline with Ollama. Scales out to your own Modal account for GPU when you need it.

▸

Human-in-the-loop

A confirmation checkpoint before anything slow, costly, or irreversible, including every wet lab submission.

Orchestrated tools

Only tools with wet lab receipts.

fova drives the best design, structure, and knowledge tools available, and shows the model structured outputs instead of raw stdout.

Design · binders · antibodies · enzymes
BindCraft RFdiffusion RFdiffusion2 ProteinMPNN LigandMPNN RFantibody BoltzGen
Structure & scoring · predict · ipSAE · filter
ESMFold Boltz-2 Chai-1 ipSAE metrics filter
Knowledge · free · no keys
Europe PMC OpenAlex Semantic Scholar bioRxiv Crossref UniProt RCSB PDB InterPro BLAST web search web fetch local PDFs Bleve corpus
Wet lab · close the loop
Adaptyv Foundry targets search cost estimate experiment status results webhook

Get fova

Install in one line.

One static binary with no runtime and no dependencies. Pick whichever way suits you, then run fova.

▸ install.shmacOS · Linux
$ curl -fsSL https://alvarogonjim.github.io/fova/install.sh | sh
▸ from sourceGo 1.22+
$ git clone https://github.com/alvarogonjim/fova && cd fova && make build

Or download a prebuilt binary

macOS Apple Silicon arm64 .tar.gz macOS Intel amd64 .tar.gz Linux x86-64 amd64 .tar.gz Linux ARM64 arm64 .tar.gz

▸ verify the install  ·  $ fova --version  →  fova 0.5.0

Close the loop

Design. Test. Learn.
Then do it again.

fova
Download fova Star on GitHub