cryocore

BioSymphony CryoCore

CryoCore helps AI agents choose, call, and chain tools for cryo-electron microscopy (cryo-EM). It combines a tool knowledge base, callable scripts, workflow instructions, and checks for passing results between tools. Use it for processing, structure analysis, validation, and rendering.

Your agent supplies the execution environment. CryoCore supplies tool records, skills, command helpers, and checks for each handoff.

What Your Agent Gains

Tool records | Agent skills | Callable tools | Workflow modules | Validation checks

Tool Knowledge Base

The registry contains more than 100 tool and resource records. It records roles, source links, versions where known, runtime requirements, license terms, smoke commands, and expected outputs. Your agent can compare options and identify missing requirements before making a call. A smoke command checks availability; processing needs a task-specific invocation.

Each record carries its own readiness and licensing status. The September source review tracks recent AI methods and releases; the adoption policy defines what a candidate needs before execution.

Tool Calling And Chaining

The agent reads the relevant skill, selects an available tool, and calls it through its terminal or API tools. CryoCore's modules describe the inputs, stages, and required outputs. The agent checks each result before using it in the next call.

The tool-use guide shows concrete calls and how to connect them. Preserve input identifiers, file formats, units, and coordinate frames between steps. Validate AI-generated maps and models against original experimental evidence before interpreting added detail.

Quickstart

Point your coding agent at skills/cryocore/SKILL.md, or follow the skill installation guide. To run the local demo, use Python 3.10 or later:

python3 -m venv .runtime/venv
. .runtime/venv/bin/activate
python3 -m pip install -r requirements-dev.txt
make demo-local

The demo fetches public RCSB metadata and coordinates, computes chain contacts and ligand neighborhoods, and creates SVG figures. Open .runtime/t2r14-open-dossier/artifacts/report.html to inspect the results. See the static preview for an offline example or the quickstart for output details.

Agent Prompt

Replace [my task] with your goal. The agent quickstart uses the same prompt.

Use CryoCore for [my task]. Read AGENTS.md, skills/cryocore/SKILL.md,
docs/tool-use-and-chaining.md, and the relevant software registry records.
Choose tools and explain why they fit. Build a sequence of calls with explicit
inputs and outputs. Run supported calls within my authorized scope, and check
each output before passing it to the next tool. If a tool is unavailable,
identify the missing requirement and continue independent steps.
Use paid compute, raw downloads, and gated tools only with explicit authorization.
Keep private data, secrets, logs, heavy data, weights, and license files out of
git and public outputs. Report the calls made, results, checks, and limitations.

Status

Pre-alpha. Support varies by tool:

Scientific software runs in separate environments. Keep private data, secrets, raw data, maps, model weights, and heavy outputs outside git. Paid compute, raw downloads, and gated tools require explicit operator authorization.

To check this checkout, run make release-check REQUIRE_GITLEAKS=1 with Gitleaks installed. Repository checks validate code and contracts; scientific conclusions require results from actual runs.

Further Reading

• Recipes and workflow modules: compose a task.
• Public data APIs: find inputs and metadata.
• Compute backends: local, cloud, and HPC execution requirements.
• Demo gallery: additional examples and outputs.
• Tool terms, data policy, and troubleshooting: resolve runtime and data constraints.
• Contributing and public release checks: maintain the toolkit.

Tracker integration is optional. CryoCore can be used directly by a terminal agent. Prediction and design belong to Structure Factory; see the ownership map.