Skip to content

Shipping today on 127.0.0.1

Nine capabilities. One dashboard.

Tandem is the layer between your keyboard and the model — multi-project orchestration, real-browser automation, an Obsidian vault for memory, and a fleet dashboard that actually knows what your agents cost. Your data, your machine.

01 / 09

Task & project management

Markdown tasks, kanban, dependencies, retries, budgets — every project tracked in one place.

  • 7-stage pipeline (inbox → planning → human_turn → active → review → paused → done) drives every task file under tasks/*.md.
  • Hard task dependencies (blocked_by / blocks) plus auto-escalation: when attempts hits max_retries, status flips to paused and the agent is cleared.
  • Per-task token budgets enforced at runtime — a full PaperclipAI-style issue surface without renting Linear, Notion, or Jira.
02 / 09

Terminal & shell

Real native shell, tmux-persistent, multi-pane, 22 themes — your terminal lives where your work lives.

  • WebSocket terminal server runs a real login shell; tmux persistence means closing the browser does not kill the job.
  • Multi-pane grids (2 / 2×2 / 3×4 / 4×4) with drag-drop reorg — the same xterm.js + WebGL stack desktop competitors charge for.
  • 22 themes shipping today — Dracula, Gruvbox, Tokyo Night, Catppuccin Mocha and more, defined once.
03 / 09

Agent orchestration

Named agents, per-agent ACL, worktree isolation, atomic claim — orchestration that survives a real fleet.

  • Roster declared in agents.json: COO, Architect, Executor, Verifier, Code Reviewer, Browser CDP, and more — each with its own prompt, allowed_tools, allowed_dirs.
  • Atomic task claim (single-flight against SQLite); writing agents are forced into isolated worktrees and auto-snapshotted on completion.
  • Live heartbeat + per-agent chat UI; chat panes route through the same tmux substrate, so agent stdout is yours to read anytime.
04 / 09

Security & governance

Localhost-only by default, approval inbox, append-only audit trail — every action is reviewable.

  • All services refuse to bind 0.0.0.0; rm/find -delete outside /tmp is blocked; audit-trail tables have no DELETE endpoint by design.
  • Human-in-the-loop approval inbox gates every agent action that touches money, the network, or disk outside its allowed_dirs.
  • Deterministic audit log: every task carries an Activity Log and links back to its session — a forensic trail you can replay.
05 / 09

Browser & web automation

Real Chrome over CDP, your real logins, multi-daemon parallelism — automation against the live web.

  • Speaks Chrome DevTools Protocol against your actual logged-in Chrome — no stealth-browser fingerprint games.
  • Multi-daemon mode runs N CDP daemons in parallel for fan-out scraping; every fetched byte hits the safe-scrape scanner first.
  • Per-site skill files (selectors, login flows, recipes) load by site, surfaced with loadouts and scope.
06 / 09

Knowledge & vault

Markdown is the source of truth — Obsidian vault, wiki ingestion, daily notes, auto context.

  • The whole vault root is your Obsidian workspace; tasks are markdown files, not DB rows — instantly visible both ways.
  • raw/ → wiki/ ingestion pipeline lands, classifies, and compiles flat wiki pages with YAML frontmatter.
  • Daily notes auto-append, weekly review workflow, and auto context management — agents read their own working memory between turns.
07 / 09

Integrations

Slack, Discord, Telegram, WhatsApp, webhooks, cron, CRM, local LLM — the connectors are in the box.

  • Social and messaging connectors ship in-tree, no marketplace round-trip required.
  • Webhooks + scheduled jobs let external events drive your agent fleet, and your fleet drive external systems.
  • Bring-your-own-key for every AI provider, plus 4 local LLM providers so inference stays on your box.
08 / 09

Dashboards & observability

Cost preview, spend-per-hour, live fleet status — the dashboard that actually knows what your agents cost.

  • Per-task and per-hour spend tracking, surfaced before you run and tallied after.
  • Live fleet status, heartbeats, and session history across every project on one screen.
  • Goal tracking with streaks so the work compounds instead of resetting each session.
09 / 09

Deployment

Linux (ARM) available now — AMD, macOS, and Windows coming soon. Run 100% local, or add an optional hosted workspace.

  • Native desktop install (Python + Next.js + Tauri) — Linux (ARM) today; AMD, macOS, and Windows coming soon. Runs entirely on your hardware.
  • Optional hosted workspace under Pro for when you can't run a local box (coming soon).
  • Mesh peer fan-out distributes work across your own machines — your boxes, your network.

Run a real fleet — on your machine.

No free tier, no trial wall. Pro is $100/seat/mo; Founding 1000 locks $50/seat/mo for life for the first 1,000 customers.

▼ Proposed new section — comprehensive all-features list (everything above is the current /features page; nothing removed)
Everything in the box

Every capability, one window.

No add-on marketplace, no per-feature billing. 25 capabilities across agents, skills, mesh, models, privacy and observability — all shipping today on 127.0.0.1, all on your hardware.

01

Agents & orchestration

Fleet

Named agents with per-agent ACL, worktree isolation, and atomic claim — orchestration that survives a real fleet.

  • Named roster COO, Architect, Executor, Verifier, Code Reviewer, Browser CDP — each with its own prompt, allowed_tools, allowed_dirs in agents.json.
  • Atomic task claim Single-flight claim against SQLite; writing agents are forced into isolated worktrees and auto-snapshotted on completion.
  • Live heartbeat + chat Per-agent chat panes route through the same tmux substrate — agent stdout is yours to read anytime, with live heartbeats.
  • 7-stage pipeline inbox → planning → human_turn → active → review → paused → done drives every markdown task file under tasks/.
  • Hard dependencies blocked_by / blocks edges plus auto-escalation: hit max_retries and the task flips to paused, agent cleared.
  • Per-task token budgets Budgets enforced at runtime — a full issue surface without renting Linear, Notion, or Jira.
02

Skills & knowledge

Vault

Markdown is the source of truth — Obsidian vault, per-site skill files, wiki ingestion, and auto context between turns.

  • Per-site skill files Selectors, login flows, and recipes load by site — surfaced with loadouts and scope so the right context arrives just-in-time.
  • Obsidian vault The whole vault root is your Obsidian workspace; tasks are markdown files, not DB rows — instantly visible both ways.
  • raw/ → wiki/ ingestion Pipeline lands, classifies, and compiles flat wiki pages with YAML frontmatter from any dropped source.
  • Auto context Daily notes auto-append, weekly review workflow — agents read their own working memory between turns.
03

Mesh & integrations

Connected

Slack, Discord, Telegram, WhatsApp, webhooks, cron — the connectors are in the box, and your machines share the load.

  • Mesh peer fan-out Distribute work across your own machines — your boxes, your network. No rented cluster.
  • In-tree connectors Slack, Discord, Telegram, WhatsApp ship in-tree — no marketplace round-trip required.
  • Webhooks + cron Scheduled jobs and inbound webhooks let external events drive your fleet, and your fleet drive external systems.
  • Browser over CDP Real Chrome via DevTools Protocol against your actual logins — no stealth-browser fingerprint games. Multi-daemon fan-out.
04

Models & compute

Any model

Bring-your-own-key for every provider, plus local LLMs so inference stays on your box — one inbox for every model.

  • BYO key, any provider Bring-your-own-key for every AI provider — no markup, no proxy tax on your tokens.
  • 4 local LLM providers Run inference on your own hardware; the model never has to leave the box.
  • Real native terminal WebSocket shell server, tmux-persistent, multi-pane grids (2 / 2×2 / 3×4 / 4×4), 22 themes — closing the browser won't kill the job.
  • Cross-platform install Native desktop install (Python + Next.js + Tauri) on Mac, Windows, or Linux. Run 100% local or add an optional hosted workspace.
05

Privacy & governance

Local-first

Localhost-only by default, an approval inbox, and an append-only audit trail — every action is reviewable.

  • 127.0.0.1 by default All services refuse to bind 0.0.0.0; rm / find -delete outside /tmp is blocked at the hook layer.
  • Approval inbox Human-in-the-loop gate on every agent action that touches money, the network, or disk outside its allowed_dirs.
  • Append-only audit Audit-trail tables have no DELETE endpoint by design — a deterministic, replayable forensic log per task and session.
  • Zero telemetry Run local and your code can stay on your machine — or run it in our managed cloud, your call. No phone-home, no usage beacon when you run local.
06

Dashboards & observability

Visible

Cost preview before you run, spend-per-hour after, and live fleet status — a dashboard that knows what your agents cost.

  • Cost preview Per-task and per-hour spend tracking — surfaced before you run and tallied after.
  • Live fleet status Heartbeats and session history across every project on one screen.
  • Goal tracking Streaks so the work compounds instead of resetting each session.

Shipping today · zero telemetry on local · data can stay on your box, or run in our cloud