EcoSeek
Scientific Agent Environment for Ecology
EcoSeek is an open-source scientific agent environment for ecological and computational biology workflows. It combines domain-specialized language models, ecological tool pipelines, and a secure gateway to help researchers run species distribution models, analyze biodiversity data, and build knowledge graphs from their laptop or an HPC cluster.
Three ways to use EcoSeek
DIY / Community
Run everything locally with EcoCoder and EcoAgent. Free, reproducible, and fully open. Uses local Ollama inference and 30+ ecological tools. No API keys needed.
Bring Your Own Key
Use your DeepSeek API key for stronger reasoning when you lack local GPU hardware. Keys are stored locally and never leave your machine. Combine with EcoAgent tools for hybrid workflows.
Lab / Managed
Connect to lab servers, workstations, and HPC clusters through AgenticPlug. Audit trails, approval gates for compute operations, and persistent sessions. Designed for multi-user research environments.
Architecture
Components
- AgenticSeek User-facing client foundation. Provides the agent runtime, CLI, and web UI. EcoSeek adds ecological providers (EcoCoder local, EcoCoder cluster, DeepSeek BYOK) and connector discovery.
- AgenticPlug Secure gateway and connector layer. Handles authentication (GitHub identity), scoped sessions, role-based authorization, approval workflows for risky operations, and audit logging.
- EcoCoder Domain-adapted QLoRA fine-tune of Qwen2.5-Coder-7B-Instruct on ~80K lines of ecological computing code. Exposes an OpenAI-compatible inference endpoint. Runs on Ollama locally or on HPC clusters.
- EcoAgent Ecological workflow and tool layer. 30+ tools including species queries (GBIF), literature search, diversity analysis, species distribution modeling (MaxEnt, SDM), triplet extraction, taxonomy resolution, and knowledge graph construction.
Ecological Tools
EcoAgent exposes 30+ tools via an HTTP API with risk classification and AgenticPlug approval gating:
- query_species
- query_papers
- compute_diversity
- fit_sdm
- fit_maxent
- extract_triplets
- resolve_taxonomy
- build_knowledge_graph
- compute_bioclim
- classify_abstract
- query_gbif_parquet
- evaluate_niche
- predict_susceptibility
- compute_effort_bias
Read-only tools run freely. Compute-intensive tools (SDM, MaxEnt, niche evaluation) carry risk metadata and can require approval through AgenticPlug before execution.
Getting Started
DIY / Community mode
Run the EcoAgent tool server locally:
pip install ecoagent[full]
python -m ecoagent.tool_server --port 8200
# Or with Docker
docker run -p 8200:8200 -e ECOAGENT_PROFILE=ci ghcr.io/alrobles/ecoagent:0.1.0
BYOK mode
Store your DeepSeek API key locally, then configure EcoSeek:
python -m sources.keystore set deepseek YOUR_KEY
# Key stored in OS keychain or ~/.config/ecoseek/keys.json
# Never committed, logged, or sent to GitHub
Lab / Managed mode
Deploy AgenticPlug as the gateway, connect to your HPC cluster:
cd agenticplug && npm start
# Configure connectors, sessions, and approval policies
# See docs/approval-workflow.md for risky operation gates
Contact
github.com/alrobles · a.l.robles.fernandez@gmail.com
EcoSeek is developed at the Reuman Lab, University of Kansas.