Redeep is a native macOS workspace where AI agents can research, code, analyze data, operate apps, and coordinate work with user-controlled access to local files, credentials, and services.
Read, search, edit, move, test, and version workspace files with structured operations or a sandboxed shell.
Agents can inspect a repository, make precise edits, run focused tests, and use Git or GitHub without leaving the session. Destructive and external actions remain subject to the session's approval policy.
Redeepread_filelist_dirglobsearchedit_filesmove_fileundo_editsbashterminalrun_testsgitgithubfind_toolsUse a persistent Python kernel, DuckDB, semantic batch operations, and rendered charts or HTML artifacts.
Kernel variables and DataFrames persist across calls. Agents can install declared packages, query local CSV, Parquet, and spreadsheet data, and return interactive visualizations in the conversation.
run_codeinstall_packagesquery_sqlsem_dataFetch pages, search the web, drive a visible signed-in browser, and—with explicit enablement—operate Mac apps.
Browser sessions can reuse the user's existing logins, hand protected login steps to the user, and save reusable site playbooks. Computer control is skill-specialized and does not become available merely because it is listed here.
web_fetchweb_searchbrowserbrowser_sessionbrowser_navigatebrowser_tabsbrowser_login_handoffbrowser_protected_controlbrowser_readbrowser_interactbrowser_screenshotbrowser_evaluatebrowser_recordsave_browser_playbookcomputerAnalyze images, transcribe local audio, synthesize speech, and—with a configured media provider—generate or edit images and create short videos.
Image, video, and speech-generation tools depend on the selected model, an active skill, or a configured provider. Local audio transcription requires macOS speech permission and an available on-device recognizer for the chosen language. Generated images render inline; video jobs may take several minutes.
analyze_imagegenerate_mediaspeaktranscribe_audioTurn Kanban cards into isolated sessions, automate column transitions, and hand work to named agents.
Board operations leave a visible activity trail. Persistent agents can have their own persona, workspace, tools, schedules, integrations, and serial task queue.
board_listboard_getboard_createboard_updateboard_configureboard_columnboard_deletecard_listcard_getcard_createcard_group_runcard_commentcard_updatecard_queue_updatecard_movecard_deleteagent_listagent_getagent_statisticsagent_notesagent_updateagent_messagetask_scriptdeferred_replyslack_user_presenceFan work out to sub-agents, delegate to external coding agents, and run saved multi-step workflows.
Sub-agents and external agents work in isolated worktrees when they edit code. The parent agent reviews and explicitly applies returned changes; long-running delegates can be listed, steered mid-run, attached, detached, or stopped.
dispatch_subagentretrieve_subagent_reportapply_subagent_changesdispatch_externaldelegate_listdelegate_attachdelegate_detachdelegate_steerdelegate_stoprun_workflowDiscover and load bundled or user-authored playbooks only when a task needs them.
Skills can include instructions, references, scripts, and declared dependencies. Redeep ships a starter catalog and also supports project-specific skills, setup packages, and browser playbooks.
list_skillsuse_skillUse host-managed credentials and integrations without exposing secret values to the model or sandbox.
Availability depends on the user's configuration and the agent's allowlist. Google Workspace reads and OAuth access stay host-side; outbound requests and configuration changes retain their approval requirements.
add_credentiallist_integrationsadd_integrationhttp_requestget_oauth_tokengoogle_calendar_readgoogle_drive_readmessagesmessage_destinationsproactive_post_grantConnect Model Context Protocol servers and search large tool catalogs without loading every schema up front.
Redeep supports local and remote MCP servers, OAuth and header authentication, per-server trust, and namespaced tools. Connected server tools appear as mcp__server__tool and are not enumerable at site-build time.
search_mcp_registryadd_mcp_serverset_mcp_server_enabledremove_mcp_serverpropose_mcp_serverfind_mcp_toolscall_mcp_toolShare sites and reports from your Mac, and move code through a guarded pull-request workflow.
Artifacts can be listed across local and cloud routes in one inventory, shared consistently, and served locally, over a private network, or through a public HTTPS tunnel. Pull-request tools update from the recorded parent, run focused verification, serialize merges, and wait for GitHub confirmation.
publish_artifactshare_artifactlist_artifactscreate_prupdate_branchfinish_mergemerge_prmerge_queue_statuslist_interrupted_operationsreconcile_operationAsk clarifying questions, expose progress, report session usage, and reconcile interrupted side effects.
Every session has an explicit approval policy. Agents can surface a plan, ask for missing decisions, inspect model and cost information, and recover cautiously when an earlier operation may have been interrupted.
ask_usertodo_writesession_infoReusable playbooks teach agents specialized workflows without filling every conversation with instructions. Add your own globally or inside a project.
commit-messages code-review software-engineering new-app skill-creator redeepctlpr-description release-notesbrand-guidelines design-review accessibility-audit design-system-extract interaction-states-passlanding-page-copy seo social-network-posting social-postsgmail google-calendar google-workspacesearch-rank-analysis data-analysis data-mapreduce web-scraping usage-statisticspptx-deck xlsx-report docx pdfdigital-operations computer-usevoice-narrationmessagesinteractive-responses