An AI agent for researchers that keeps your data local

Research tasks it handles

Your research runs to dozens of pages; C5 hands it back as a finished report with a short summary. Ask in plain words, attach your notes, papers or data, and say what done looks like.

• Long structured reports — the factory explosion investigation sample is a 55-page report with a 4-page summary, produced on a desktop computer with local models.
• Multi-disciplinary responses — the offshore wind sample answers an environmental impact report in 24 pages plus a 9-page summary, again with local models.
• Research compilations — routine digging gathered and filtered down to the findings you need.
• Recurring digests — a scheduled roundup of a field's headlines, with web search turned on for the task.

Factory Explosion InvestigationReal C5 deliverable →Offshore Wind EIR Multi-Disciplinary ResponseReal C5 deliverable →News & Research CompilationReal C5 deliverable →News: Top 5 Al HeadlinesReal C5 deliverable →

Behind each one, C5 turns your request into a spec, staffs specialists, runs them in parallel and reworks any step that misses — see how multi-agent orchestration works.

Sources, checking and a record of the work

Every result is reviewed against your brief before it reaches you. What does not hold up goes back to be redone, and nothing half-finished is passed off as done.

• Reference lists — PDF deliverables can carry a reference list in APA style.
• Links kept with each item — material gathered from the web keeps its headline, summary and link, and an image fetched from the web is captioned with the site it came from.
• A readable record — the plan, each step, what each agent did and what came back stay open to you. Tool decisions are logged in Ops › Audit, beside audit entries that are tamper-evident.
• Signed deliverables — each deliverable is signed when it is produced, recording the workspace it came from, and the Library's Validate checks it has not been changed. Verification is done by your own C5; there is no public key for third parties yet.

Check every source before you cite it. C5's review checks the work against your brief, not against the literature, and a model can still get a fact wrong.

C5 is the wrong fit if your workflow depends on a connector to a specific literature database: none is documented, so agents work from what you attach and what web search returns.

Memory that remembers your project

A later task should find what an earlier one found without you pointing at the file. C5 keeps an index of the notes, memories and documents its agents produce so it can. It is an AI agent that works with your local data, not a chat window that forgets.

• Keyword and semantic search — find a document by its words, or by what it means when it uses different words. Semantic search uses models C5 downloads once and runs on your machine.
• Projects — keep every task, file and note for one study together in the Library, and search inside the files, not just their names.
• Your earlier AI work — C5 can index the ChatGPT Codex and Google Antigravity sessions already on your computer, and Claude sessions when the Claude app is installed on the same computer, without uploading them.

Unpublished and sensitive data

Unpublished data and participant records come with rules about where they may go. C5 is a local AI agent: its database, files and history are stored encrypted on your computer. Run a local model through Ollama, LM Studio or llama.cpp and the data stays on your machine while it is processed; with a cloud model, the prompt and material a task sends go to that provider under your own account.

• Offline — with local models and the one-time downloads in place, C5 needs the internet only for an occasional licence renewal, or not at all with an air-gapped licence. See an offline AI agent.
• Checks on what goes out — before an agent's tool call runs, C5 looks for credentials and personal details such as email addresses and phone numbers, and redacts them by default. It checks what agents send, in recognisable formats; it is not a guarantee.
• Mix models by task — a local model for the sensitive material, a cloud model for public sources.

C5 makes no claim about any ethics board's or data office's rules. Check what your protocol allows before you choose a model.

A digest of your field that runs itself

Write the brief once and make it a schedule: weekly, monthly or on a date. Each run lands on the board as its own task and its result in the Library, and follow-up recipes can run on what it produced. See how to schedule an AI agent to run every week.