A clinical documentation tool for working physicians. Pair it with a local Ollama model and chart text never leaves your laptop — no cloud, no SaaS, no GMTek server in the path. Eight modes, three providers, one workflow. Ships as a native binary; Python optional.
One paste of a chart, one click, one of eight purpose-built outputs — pasteable straight into the EHR. Every call is stateless: zero conversation history, zero cross-patient bleed. Quick-fill templates for HTN follow-up, DM annual, post-op, URI, Medicare AWV, and well-child checks get the chart in front of the model in seconds.
Paste full chart — get a "Review for Next Provider" handoff note plus an MDM Summary scored Low / Moderate / High complexity.
Paste a specialist consult note — get a single PCP action paragraph distilled to what primary care actually needs to do next.
Reads the visit, suggests the supportable E/M level (99213 vs 99214 vs 99215) with the documentation evidence. Pays for itself in a week.
Pulls candidate ICD-10 codes from the assessment with the supporting chart language for each — fewer rejections at the clearinghouse.
"Lint mode" for charts. Flags gaps that, if documented, would qualify a higher E/M code. Direct revenue impact, not stylistic nagging.
Generates a 6th-grade-reading-level handout for the visit. Optional translation. Cuts the "what did the doctor say?" follow-up calls.
Brainstorming aid only — surfaces differentials worth considering, with a prominent disclaimer. Not clinical decision support.
Five sub-types: prior auth, letter of medical necessity, specialist referral, work/school excuse, disability/FMLA support. Drafts; you edit and sign.
Most "AI for doctors" tools route chart text through a vendor's cloud. Chartwright doesn't have to. Pair it with Ollama running on the doctor's own laptop and the chart text physically cannot reach the public internet. When you do want a frontier-model, the path is honest and audited — not hidden.
Browser → local Python server → Ollama on localhost:11434. No cloud, no GMTek server. The recommended path for any clinical PHI. Whichever chat models you've pulled show up in the dropdown automatically.
Anthropic Claude or OpenAI GPT directly from your browser via the local proxy — no GMTek server in between. Sending real PHI to a cloud provider requires your own BAA with that provider. A yellow banner reminds you whenever a cloud provider is selected.
Local Safe Harbor stripper runs before any cloud request — names, dates, MRNs, phone numbers, addresses, the full 18-identifier list. Defense-in-depth, not a BAA substitute, and the doctor sees the redacted version before send.
Every request appends one line to ~/.chartwright/audit.log: timestamp, provider, model, mode, byte counts, latency, status. No chart text. No output text. The technical-safeguard audit trail HIPAA actually asks for.
API keys for cloud providers live in browser sessionStorage only — cleared when the tab closes. No file, no keychain, no leaked env var.
Passphrase-protected chart history with Argon2id KDF + AES-GCM. Search your own past work; lose the passphrase, lose the archive. Nothing about it leaves the laptop either.
GMTek is not a covered entity or a business associate. Chartwright is a local tool a doctor installs on their own machine — we never touch your charts and we don't sign BAAs. In Ollama mode that's the whole story. In cloud mode, the BAA you need is between you and Anthropic or OpenAI, not us. Most physicians prefer that arrangement once it's stated plainly.
Undercoding is the silent revenue leak in primary care. A 99213 instead of a supportable 99214 is roughly a $40 swing per visit — and the average internist gives away tens of thousands of dollars a year to it, often because the documentation that would justify the higher level just isn't on the page.
Chartwright's E/M Code Recommender reads the visit and tells you the level the documentation actually supports — with the evidence quoted back. Its Documentation Completeness Check goes a step further: it flags the missing pieces that, if you added two sentences, would push the visit into the next level. Both modes are about recovering revenue you've already earned but haven't billed for.
At $79/month — roughly a quarter of what enterprise tools like DAX Copilot or Suki AI charge per provider — the tool pays for itself the first time you upgrade two visits a month. The other six modes — handoff notes, ICD-10 extraction, patient handouts, letter drafts, differential aid, specialist summaries — come included.
Chartwright recommends; the physician decides. Every recommendation cites the chart language it's based on. Never bill for documentation you didn't write — Chartwright is a documentation and coding aid, not a billing autopilot. The point is to capture what you actually did, fully and accurately.
If you're the practice — no IT department, no enterprise EHR add-on budget, no scribe — Chartwright runs on your laptop and recovers documentation revenue you're currently leaving on the table.
Multi-doctor profiles built for shared workstations. Each clinician's history, templates, and audit log stay scoped to their profile. The Small Clinic plan covers up to five.
Chartwright was originally built for one. The stateless-per-chart design, quick-fill templates, voice dictation, and zero context bleed all serve the "I need to focus on this one chart, right now, without pollution from the last six" workflow. Still the original use case — kept on purpose.
If your compliance officer flinches at "AI in the cloud," Ollama mode is the answer. Chart text never leaves the machine, the audit log is clean, and there's no third-party SaaS storing PHI to worry about.
The E/M Recommender + Completeness Check are the fastest feedback loop on coding you'll find. See the level the documentation supports, see what's missing, learn the rules in real cases instead of CME slides.
Lemon Squeezy is the merchant of record — VAT and US sales tax handled. License keys activate the binary; no account, no SaaS lock-in. Cancel and you keep the binary; it just falls back to the free trial limits.
First 50 doctors: $49/month for Year 1 on Solo Doctor (regularly $79). Lock in early-adopter pricing during private beta. Enterprise (16+ doctors): contact josh@gmtek.ai for hospital-system pricing.
Cloud-provider API costs are pass-through. If you use Anthropic or OpenAI, you pay them directly — usually $5–$20/mo for a working solo doctor, often less. Ollama mode has zero per-chart cost beyond the electricity to run your laptop. We don't mark up tokens.
PyInstaller-built single binary for Mac, Linux, Windows. Doctors don't install Python. Double-click, browser opens to localhost:7432, you're working.
Voice-to-text via faster-whisper running entirely on the doctor's machine. Same HIPAA story as Ollama — audio never leaves the laptop.
At /analytics — charts/day, mode mix, time-saved estimate, error rate. Built entirely on the audit log. Never on chart text.
Server-Sent Events stream tokens as they generate. The doctor sees output forming in real time — no 30-second blank screen on long responses.
HTN follow-up, DM annual, post-op, URI, Medicare AWV, well-child check. One click, scaffold appears, fill in the specifics, run.
Every chart review is a fresh API call with no conversation history. Patient A never bleeds into Patient B. Zero cross-contamination, by design.
Chartwright is in private beta with practicing physicians. The next-tier rollout is for solo PCPs and small clinics who want documentation help without sending PHI to a SaaS. Email us; we'll send a binary, a license key, and the install guide.