A fleet, A fleet, not a chatbot.not a chatbot.

Reads the prompt and classifies intent, complexity and entities.

Breaks a task into ordered steps, each tagged with the kind of specialist it needs.

Matches every step to the best-fit agent and the right depth of reasoning.

Runs the plan step by step, streaming each event so you can watch the work happen.

Reviews the finished result against the brief and asks for a retry when quality is short.

Pulls relevant memory, prior work and uploaded files into context before the model sees them.

Speaks the open agent-to-agent protocol so external agents can ask askFinz to do real work.

Plans, browses and synthesises a long-form study, streaming every step.

Runs the same prompt across several minds in parallel, then a judge merges the strongest drafts.

Generator + reviewer pair — scores a draft against a rubric and sends it back for one revision.

Reads N sources across M questions and fills a comparison grid with a confidence score per cell.

Pulls inbound citations for any source and labels each one as supporting, contradicting or neutral.

Lint, gap analysis and rubric scoring on a draft — surfaces uncited claims and missing sections.

Licences public-library documents for training, settles royalties and tracks owner credits.

Reads any document and proposes a reusable structure you can apply to the next study.

Turns tracked-change proposals into accept/reject decisions with an audit trail.

Pairs in the editor — reads the repo, edits files, runs commands, answers in your terminal.

Highlight a span of code and ask for the rewrite right there; preview the diff before accepting.

Run several code agents at once on the same prompt or on independent slices of a task.

Lets the main agent spawn sub-agents on its own and weave their answers into one reply.

Drops a focused agent onto a sub-task with its own scratchpad, then hands the result back.

Multi-agent planning — parallel planners propose approaches, the best parts get merged.

Runs thousands of prompts through a model in one cheap overnight job and streams results as ready.

Reusable role packs (refactor, test, review, debug, security, optimise) the agent loads on demand.

Fires shell commands before, during or after a tool call so your project rules stay enforced.

Exposes the workspace as a tool registry so other agents — even ones outside askFinz — can call it.

Cursor presence, real-time cursors and shared selections when more than one person is in the file.

A specialised mode that hardens its prompt for reproducing, isolating and fixing nasty bugs.

Reads broadly, cross-references claims, cites sources and structures the answer.

Generates, reviews, debugs and refactors code in any major language.

Drafts, edits and rewrites to a target tone — clean structure, plain language.

Works with numbers — tables, comparisons, charts and quantitative reasoning.

Distils long content into the essentials with a structured outline and TL;DR.

Secondary mind running alongside the primary, surfacing alternative suggestions as you go.

Reaches into hundreds of publishers, extracts full articles and keeps the index fresh.

Sentiment, bias and category — every story scored so the briefing reads honestly.

Builds the hourly transcript across today's top stories and renders it as audio.

Scans the headline window for sector and ticker risk signals over a rolling fourteen-day view.

Drives a real, headed browser to read pages, follow flows and fill forms — with a screencast.

Rotates clean exit-IPs and degrades gracefully so a brittle site doesn't end the session.

Reads every workspace and keeps the askFinz vector index in sync as you work.

Sweeps the index for stale, broken or low-quality entries and quietly heals them.

Executes a custom flow you've designed — connects steps, ships the output, logs the run.

Saves the facts you ask it to remember and pulls them back into future conversations.

Shares a project-scoped memory across collaborators with last-write-wins reconciliation.

Once a week, distils recent session logs into durable project memory you can review.

Semantic recall across everything memory has ever held — surfaces the right fact, fast.

Per-tool approval rules with wildcard allow/deny so risky actions always pause for you.

Queues every file write for inspection before it lands — accept, reject or rewrite the patch.

Restricts which files and domains an agent may touch, even when it is told to do more.

Cross-checks an action against your stated intent before money is spent or content is sent.