Platforms & Systems
Applications
Clinical Workflow Orchestration
Multiple agents work through layered planning, execution and verification, with a full audit trail.
MDT Decision Support
For complex cases: prepares the case before the meeting and gathers specialty opinions during it.
MedHistory Agent
Organizes the records of one inpatient stay; answers are traceable: click a fact ID to open the source note.
AI Output Verification
Checks the sources AI outputs cite; currently covers references: whether each exists, whether its metadata has been tampered with, and whether it has been retracted.
Capabilities
Terminology Normalization
Maps variant names of diseases, lab tests and drugs to standard terms.
Derived Medical Events
Derived from facts in the clinical record using rules; the evidence is traceable to the source text.
Models
Specialized Models & Compound AI Systems
Training progress, evaluation results and testbeds for specialized models.
Data
Clinical Data Annotation & Quality Control
Includes dataset storage; accounts are provisioned by the administrator.
Medical Knowledge Harvesting & Monitoring
Runs continuously and can be reconfigured when a source site is redesigned.
Research
Recent Updates
Service access consolidation: all systems, APIs, and API docs have switched to direct intranet addresses; requests are handled entirely within the local network; the portal is for introduction only.
MedGo Extract upgraded: the front rules layer, extraction and post-processing are tuned end-to-end on in-house records, with merge parameters refined. The rules layer was regressed on the full outpatient corpus.
DeID model upgraded with on-premises fine-tuning: retrained, with clearly better key-field recognition. Recall: name 0.98, ID 0.97, phone 0.98, organization 1.00; overall F2 0.97.
Deep Research: optimized for the Chinese-language interface.
/plugin install deep-research@medgo-skillsLights performance refactoring: with the status of dozens of sessions monitored in real time, steady-state CPU usage is down from 20–79% to about 2%; status-light layout and visual design refined, with the indicator anchored to the input area; download Lights.dmg.
MedGo Extract updated to V1.3: assertion and medication-status fixes — terms that contain 无 (wú, “no/without”) as part of the word, expressions such as 未除外 (“not ruled out”) and 不建议 (“not recommended”), trailing discontinuation cues and time-specific doses are now recognized more accurately.
ChatRecall 4.4.2 released: faster long-conversation retrieval; OpenCode / MiMo sessions indexed; 40,000 sessions and 2.1 million messages.
Engineering standards upgrade: version-control audit completed across all projects and live services; each is being brought under Git, with code committed and data and models kept out of the repositories.
Access-layer upgrade: the gateway and cryptographic libraries updated to the latest patch releases; all subdomains are now under access control; version info removed from response headers.
Site-index optimization: search coverage expanded; search keywords and the index page updated in sync.
MedHistory Agent interface upgraded: retrieval and verification are shown live; each fact in an answer links back to its source in the chart (document, date, author and original excerpt).
Pulse system new dashboard launched: the collection dashboard has been redesigned to bring together multi-source monitoring, incremental ingestion, PDF archiving and self-healing status, with key metrics updating live.
Pulse system: more public data sources connected, with continuous testing of collection stability, new-content detection and rule repairs (applied automatically once validated, escalated to a human on failure); end-to-end telemetry can be followed on the dashboard.
MedHistory Agent now autonomously verifies before it answers: multi-step retrieval across labs, medications, clinical problem timelines and chart records; every citation and value is checked against what was retrieved, mismatches trigger a re-query, and each step is shown live.
Term Normalization Dictionary version 20261021.1: mismapping and granularity fixes are live; the API now returns dict_version.
Lexicon standalone page is live: an introduction to the standard terminology resource, covering its scale, sources and terminology quality.
MedHistory Agent launched: load a case and ask follow-up questions about trends in clinical measurements, medication changes and unresolved problems; answers rely only on what the chart records and cite their sources.
Term Normalizer updated to V2: adds disease concept sets to determine whether a diagnosis belongs to a given disease category (e.g. coronary heart disease), returning the result and its rationale.
MedGo Extract updated to V1.2: bug fixes and stability improvements, with the API documentation updated accordingly.
ChatRecall now integrates the OpenCode and MiMo session stores, with unified search and usage tracking.
Concept Normalization review updated: Chinese-language literature published in China added; Section 7 gains three new parts on coding hierarchies, mapping between the National Clinical and Medical Insurance editions of ICD-10, and Chinese terminology resources.
Concept Normalization review and field guide updated: every citation re-checked sentence by sentence against the source, wording and data definitions refined accordingly, an abstract added, and both documents revised together.
Concept Normalization field guide published: distilled from the full review and organized by the stages of building a pipeline, covering terminology choice, candidate generation, LLMs and abstention, and the use of hierarchies in rules.
MedGo Skills launched: AI agent skills distilled from day-to-day work, with one-command install on the intranet.
Concept Normalization review published: an overview of synonym mapping, terminology hierarchies and quality evaluation in clinical concept normalization.
Hallucheck Verification engine upgrade: on external benchmarks, every problematic citation is detected with zero false positives on genuine references; under the same protocol, its decision accuracy ranks first among six comparable tools.
Clinical Extraction review published: a comprehensive overview of methods, architectures and evaluation for information extraction from Chinese medical records in the era of large language models.
Record standardization API now available: structuring and normalization in one call.
Lab normalization API now available: maps terms from clinical records to in-house standard names and concept IDs.
MedGo Extract updated to V1.1: more complete fact packs and fewer missed extractions.
Record de-identification API now available: returns the masked text and the position of each masked span, using multi-pass LLM detection and review.
Record structuring API docs published, bring your own schema: submit the JSON structure you need and get results in that same structure, with every value traceable to the source text.
ChatRecall 4.4 released: the multiple copies created when a session is resumed are now merged into a single result, cutting duplicate entries in search results from nearly 30% to about 1%.
History Summary API now available: submit all documents for a patient in one request; the API returns a history summary organized into sections, with every sentence traceable to the source text.
MedGo Extract updated to V1: adjusted based on post-launch user feedback and testing on real cases.
MedGo Extract API now available: submit all documents for a case in one request; the API returns a unified fact list, with each fact's assertion status, attributes and location in the source text.
Hallucheck technical report updated: external validation on multiple public benchmark datasets, with expanded related work and supporting literature.
Portal adds “Ask this site”: retrieval over the site-wide knowledge base (RAG), multi-turn dialogue, sentence-level citations, with built-in hallucination guardrails and layered prompt-injection defenses.
Agent Memory review published: a comprehensive overview of research on memory mechanisms in LLM agents, written in Chinese and running to more than 25,000 characters.
Hallucheck API now available: reference verification can be integrated directly into applications; a single request returns existence status, bibliographic comparison results and retraction status.
MedGo Extract improved: following a review of over 10,000 papers, the temporal layer has been rebuilt so the model only classifies while the code computes the dates.
ChatRecall 4.2 released: in A/B tests on real conversations, a single call to the new API retrieves about 90% of the key points (vs. about 30% before) and cuts time per question by about 20%.
Orchestrator clinical workflow orchestration examples refreshed: every example now executes against the live agents and services on the server.
MDT sub-feature Hospital Course Digest released: pre-meeting hospital-course review that aligns each clinical problem timeline with medication starts and stops in one overview table, traceable to the source text.
MedEvents Derived Medical Events released: checks each record fact against rules and produces derived events with evidence chains.
CaseHub adds History Summary: two simulated cases show the path from source text to summary.
Model Studio adds the PreLabel engine.
ChatRecall 4.0 released: conversation-history search returns more than ten candidate sessions with their original text in one call.
Portal launched: site-wide search covers every system's pages, sections and API docs.
