tars

README

TARS

Homepage: https://tars.marvin-42.com — project overview, features, and quickstart.

TARS is a local AI agent runtime that runs on your machine, under your control.

TARS runs as a single Go binary on your machine. From the browser console you can directly inspect and control its work — chat with dockable Git Inspector and tool panels, message-level session forks, an Agent Runtime page with list/tree/Gantt/Flow views, Memory Inbox review, scheduled jobs, a background watchdog, and nightly reflection. Everything is configurable via YAML and extensible via skills, plugins, and MCP servers. Telegram and webhook channels are available when you're away from the browser.

The name comes from the TARS in Interstellar — practical, direct, dependable when things get complicated. TARS aims for that. Not affiliated with the film; the name is borrowed.

Comparison

	OpenClaw	Hermes Agent	TARS
Language	TypeScript	Python	Go (single binary)
Primary UI	CLI	CLI + Agent Runtime API	Browser console (CLI / Telegram / webhooks also)
Sub-agents	ACP + subagent runtimes, push-based completion, Docker sandbox	ThreadPoolExecutor (max 3), ephemeral prompt, credential override	Agent Runtime executor with per-task model tier, allowlist policy, depth control
Model routing	Per-agent model override	Per-child provider/model override, MoA (4 frontier models)	3-tier named bundles (heavy/standard/light) with role→tier config mapping
Memory	Session transcripts	Honcho/Holographic plugin hooks	Durable Markdown memory + semantic search + review-before-store extraction + nightly reflection
Background	None	None	Pulse watchdog (1-min) + Reflection nightly batch
Scheduling	None	None	Session-bound cron jobs with audit logs
Channels	CLI	CLI + Agent Runtime API	Console + Telegram + webhooks
Context mgmt	Per-session	ContextCompressor (50% threshold, protect-last-N)	Structured compaction with identifier preservation + light-tier LLM summary
Extensibility	Built-in tools	Toolsets (terminal, file, web, delegation)	Skills + companion CLIs first, plus gated plugins/MCP from Skill Hub

Key Features

Chat + Memory

The primary interface. Browser-based console at http://127.0.0.1:43180/console.

• Multi-session chat with full LLM tool-calling loops and transcript snippet matches in session search
• Session lineage metadata and stable transcript message IDs power message-level chat forks: hover a persisted user or assistant message, choose "Fork from here", and continue in a child session that reuses setup, plan state, tool policy, prompt override, and workspace roots through that point
• The Console Lineage page renders root and forked sessions as a git-log-style tree with fork point previews, click-through navigation back into chat, and reviewable fork insight promotion into Memory Inbox without mutating the parent transcript
• Tool calls render as collapsible rows with live elapsed time, compact argument previews, and automatic error expansion
• @ file and directory mentions from the session Files roots for explicit context injection
• / command autocomplete for built-in chat actions and explicit user-invocable skill selection
• First-turn tier recommendation suggests heavy, standard, or light before the first expensive chat call; accepting or overriding it routes that turn through the chosen tier and records the outcome for later usage analysis
• /config opens advanced per-session tool, skill, automation consent, and TARS style controls when a selected session needs explicit overrides; permission changes show a deterministic risk/impact preview before saving
• Session Health badge and dockable recommendation panel flag long context, stale plans, broad permissions, noisy prior memory, and idle sessions with direct actions for compacting, task review, config trimming, prior-context review, or skill extraction
• Prior Context preview panel showing the exact memory section, source badges, and token budget for the current draft
• Chat panels share a Dock Manager: Sessions and tool panels, including the Git Inspector, can move between left, right, bottom, and fullscreen zones, with drag-resized dimensions persisted in the browser
• Files workspaces launch the integrated shell in the bottom dock at the selected root or browsed subdirectory, while keeping a macOS Terminal fallback available
• Workspace file previews use workspace storage by default, while selected filesystem roots are served through the explicit filesystem files API boundary
• Durable memory: MEMORY.md, reviewed experiences, daily logs, semantic embeddings
• Memory Inbox review queue for approving, rejecting, or merging reflection-derived memory candidates before they enter durable recall
• Fresh workspaces omit legacy KB Wiki scaffolding while preserving any existing memory/wiki files
• Editable memory assets plus Tool path and Prefetch path recall tests through the console/API
• Structured transcript compaction preserving message identifiers and recent context
• System prompt customization via USER.md, IDENTITY.md, AGENTS.md, TOOLS.md

Sub-Agent Orchestration

Spawn read-only agents for research, planning, and specialized tasks:

# workspace/agents/explorer/AGENT.md
---
name: explorer
tier: light
tools_allow: [read_file, list_dir, glob, memory_search]
---

Use subagents_run when tasks are independent and can fan out in parallel:

{"tasks": [
  {"prompt": "find all API endpoints", "tier": "light"},
  {"prompt": "design the migration plan", "tier": "heavy"}
]}

Use compare mode when 2-3 safe subagents should inspect the same prompt independently and TARS should keep their outputs side-by-side:

{"mode":"compare","tasks": [
  {"agent": "explorer", "title": "Explorer pass", "prompt": "find the root cause"},
  {"agent": "reviewer", "title": "Reviewer pass", "prompt": "find the root cause"}
]}

In Console Chat, subagents_run renders as a progress card with running/completed/failed counts, elapsed time, compact task titles, and direct links to each Agent Runtime run once they are available. Compare-mode results add common findings, conflict candidates, sourced evidence snippets, and side-by-side output panes.

Advanced staged-flow tools are available only when explicitly allowed for a session: subagents_orchestrate runs dependency-aware parallel / sequential steps, and subagents_plan uses the heavy-tier planner model to draft such a flow. When staged-flow tools run from chat, TARS mirrors the generated plan and live step lifecycle into the session Tasks panel so the right rail shows pending, in-progress, completed, and cancelled subagent work.

Experimental consensus mode remains hidden from the default subagents_run schema unless agentruntime.consensus.enabled is explicitly set.

Tier resolution priority: task tier > agent YAML tier > config default.

The Console Agent Runtime page exposes Runs | Subagents tabs. Use Runs to filter execution history by status, time range, and prompt text, switch between list/tree/Gantt/Flow run views, pan and zoom an interactive Svelte Flow graph, jump back to the originating chat session, scan today/7d/plan cost totals, scrub timestamped run events with Replay, restart failed runs from prompt/failure checkpoints with optional agent, tier, model, or prompt adjustments, inspect each run's cost/token flow, review file attention for frequently read or edited workspace paths, and open the git diff timeline that attributes workspace file changes to the run, session, agent, and plan step that produced them. Use Subagents to inspect the active catalog, default/effective LLM tier, resolved provider/model, source file or command entry, tool policy, and recent run links. Workspace AGENT.md profiles can update their default tier, draft new subagents with an LLM-assisted builder, generate run-derived profile recommendations with preserved provenance, preview and approve LLM edits or recommendations, and archive inactive workspace profiles from this detail view.

3-Tier Model Routing

Route workloads to different models for cost and quality optimization:

Tier	Purpose	Example
heavy	Planning, complex reasoning, architecture	claude-opus-4-6, gpt-5.4
standard	General chat, agent loops, tool calling	claude-sonnet-4-6, gpt-5.4
light	Summarization, classification, pulse, reflection	claude-haiku-4-5, gpt-4o-mini

# tars.config.yaml
llm:
  providers:
    default:
      kind: anthropic
      auth_mode: api-key
      api_key: ${ANTHROPIC_API_KEY}
  tiers:
    heavy:
      provider: default
      model: claude-opus-4-6
    standard:
      provider: default
      model: claude-sonnet-4-6
    light:
      provider: default
      model: claude-haiku-4-5
  default_tier: standard
  role_defaults:
    pulse_decider: light
    agentruntime_planner: heavy

Each system role (chat, pulse, reflection, compaction, agent runtime agents) maps to a tier. Background surfaces default to light, keeping costs low. Console Chat also classifies the first user message before the first expensive call and can route that turn to heavy, standard, or light based on the accepted recommendation or user override. Session style controls add directness, humor, caution, and autonomy values to the chat prompt while keeping autonomy bounded by explicit consent and approval policy. If advanced staged subagent planning is explicitly enabled for a session, llm_role_agentruntime_planner is exercised by subagents_plan; TARS logs the resolved role, tier, provider, model, and source for chat and agent runtime LLM calls so tier selection is traceable in runtime logs. The Console Settings page includes a typed llm.tiers editor for adding, renaming, editing, and removing tier bindings without hand-editing JSON.

Background Surfaces

Two isolated surfaces run independently from user chat:

• Pulse — 1-minute watchdog scanning cron failures, stuck runs, stalled chats, disk pressure, Telegram delivery health, and reflection status. LLM classifier picks ignore / notify / autofix. The Console renders recent signals as incident cards with likely cause, evidence, recommended action, safe navigation, and re-check controls. Autofixes are whitelisted in config, and stalled-chat continuation requires per-session auto-resume consent.
• Reflection — Nightly batch (default 02:00–05:00) running memory reflection (Memory Inbox candidate extraction) and stale empty-session pruning.

Both use the light tier by default and have no access to user-facing tools (enforced at compile time via RegistryScope).

Scheduling

Native cron with session binding:

• Cron expressions and one-shot at: schedules
• Session-bound jobs inherit the session's tool policy, work dirs, and prompt override
• Audit logs: artifacts/<session_id>/cronjob-log.jsonl
• Console Cron page for global job management plus per-session Cron tabs in chat context

Channels

Multi-channel I/O beyond the web console:

• Telegram — Bidirectional messaging with pairing-based access control
• Webhooks — Inbound HTTP triggers for external integrations
• Assistant — macOS popup and voice helpers that share the core ~/.tars/workspace default unless overridden
• Local — Direct API calls for scripts and automation

Extensibility

TARS favors on-demand extension over always-resident tool registrations. Domain-specific capabilities are shipped as skills (plus optional companion CLIs) from the Skill Hub rather than compiled into the TARS binary — this keeps the chat system prompt small no matter how many capabilities a user installs.

• Skill Hub — Public registry of skills, plugins, MCP servers, and domain packs. Install individual packages with tars skill install <name>, tars plugin install <name>, tars mcp install <name>, or install a reviewed bundle with tars pack install <name>. Skill, plugin, MCP, and pack member installs/updates are first materialized in a temporary workspace and sandbox-validated before the real workspace/skills/<name>/, workspace/plugins/<name>/, workspace/mcp-servers/<name>/, or skillhub.json changes. Hub entries can publish quality metadata such as score, last update, passing tests, required tools, permissions, companion CLI presence, and install count; the Extensions console renders these trust signals before install. Update commands report updated, skipped, and failed entries so failed package refreshes are visible. The hub is the first place to look before writing a new capability, and the only place to publish one.
• Skills — Markdown instruction files (YAML frontmatter + body) with optional companion scripts. A skill's frontmatter can set recommended_tools: [bash], slash: /name, aliases: [...], and smoke_tests: [...]; users can invoke eligible skills directly from chat via /name autocomplete. Companion CLIs keep their interface out of the system prompt until the skill itself is picked. The Chat Skill Inbox can extract reusable skill candidates from a session, preserve provenance/evidence, and save approved candidates as local workspace/skills/<name>/ drafts. The Extensions console can also draft, sandbox-test, and save local skills before you publish them to the hub, and shows sandbox pass/fail reports for hub skill installs. See daily-briefing in the hub for the canonical pattern.
• Plugins — Advanced packages that bundle skill directories and optional MCP server declarations with manifest metadata and runtime gating. Plugin-declared MCP servers are disabled by default and require extensions.plugins.allow_mcp_servers: true before they can launch. Built-in Go plugins and plugin HTTP routes are not an active extension surface.
• MCP — Configured, hub-installed, or plugin-declared local stdio and remote HTTP/SSE/WebSocket servers with bearer or OAuth auth. Use MCP for third-party integrations that cannot be expressed as a CLI the bash tool can call. The Extensions console can draft, stdio-test, and save local workspace/mcp-servers/<name>/ MCP packages before you publish them to the hub.
• Browser — Playwright-based automation for web interaction (shipped as a hub plugin).
• Per-project overrides (.tars/) — Drop a .tars/settings.json (team-shared, commit it) or .tars/settings.local.json (per-user, gitignore it) under the chat session's active working directory and TARS will fold those values into the session's tool_config / prompt_override for chats run in that directory. A sibling .tars/skills/<name>/SKILL.md adds project-only skills, and .tars/commands/<name>.md registers thin slash-command aliases over existing skills via a target_skill: frontmatter key. The active cwd is shown as an amber chip in the chat header (or via /cwd); merged values render with a shared / local source badge in the Session Config panel. The whitelist is narrow on purpose — credentials, hooks, and arbitrary binary paths are blocked at load time. See CLAUDE.md for the full schema.

The Extensions console validates MCP draft names before requesting generated files, so incomplete drafts stay local to the form instead of sending avoidable server requests.

When to build a hub skill vs. a core feature: if the capability is domain-specific (one site's logs, one vendor's API, one workflow), it belongs in tars-skills as a skill + CLI. Builtin tools inside this repo are reserved for universal surfaces (file ops, memory, agent runtime, channels) that every session uses.

Install

Homebrew:

brew tap devlikebear/tap
brew install devlikebear/tap/tars

Curl:

curl -fsSL https://raw.githubusercontent.com/devlikebear/tars/main/install.sh | sh

Quick Start

# Initialize workspace and config
tars init

# Start the server (it boots in setup-only mode if no LLM is configured yet)
tars serve

# Open the web console — the onboarding wizard runs automatically when needed
tars

On first run, if ~/.tars/config/config.yaml has no llm_providers / llm_tiers, the server boots in setup-only mode: only the wizard endpoints + /console are active and tars api serving on … (setup-only mode) is printed to stdout. Open http://127.0.0.1:43180/console, the SPA detects needs_setup=true from /v1/healthz, and walks you through three steps: pick a provider (kind + alias + api_key), bind heavy/standard/light tiers to a model, then save & restart. The wizard is also re-runnable later from the Settings page (/console/config → "설정 마법사 다시 실행 →") if you need to swap providers or tweak tier bindings.

To skip the wizard and edit by hand, set credentials directly and validate:

export ANTHROPIC_API_KEY="your-key"   # or OPENAI_API_KEY, etc.
# Edit ~/.tars/config/config.yaml under llm.providers / llm.tiers
tars doctor --fix
tars serve --config-check              # exits 1 in setup-only mode

On macOS, tars service install && tars service start manages tars serve as a LaunchAgent. Once a plist exists, tars service start, stop, and status keep working from the plist and launchctl state even if the config needs repair.

When TARS is exposed behind a reverse proxy path, pass that base path in --server-url; CLI API calls and the console opener resolve /v1/* and /console from the same base.

For local console development, set TARS_CONSOLE_DEV_URL=http://127.0.0.1:5173 while Vite is running. The dev proxy keeps Vite assets, HMR, and favicon requests mounted under /console.

Console Pages

The sidebar keeps Home on the TARS logo and groups the working pages into Work, Operate, and Setup.

The sidebar footer keeps server, Pulse, Reflection, and active session status visible with 30-second refreshes and direct jumps to each detail page.

Frontend API response types are maintained in the console source with the current contract policy documented in docs/frontend-api-types.md.

The console header includes an EN/KO language toggle. The selected locale is stored in browser localStorage as tars_console_locale; first load falls back to navigator.language and then English.

Set usage.limits.daily_tokens (or usage_daily_token_budget) to show a daily token budget chip in the console header. The chip uses UTC day boundaries, counts input plus output tokens, hides when set to 0, and links to today's analytics focus once usage reaches the error threshold.

The Chat header shows the active plan goal, completed/total task count, and a Session Health badge without opening side panels. First-turn Chat prompts can show a cost/quality tier recommendation card before the first LLM call, and the Context HUD shows the selected tier plus whether the recommendation was accepted or overridden. The dockable Chat Health panel recommends compacting, splitting/fork review, task cleanup, permission trimming, prior-context review, or skill extraction when the session becomes long, stale, over-permissive, noisy, or idle. The dockable Chat Contract panel keeps the active work contract explicit with goal, scope, done criteria, verification commands, expected artifacts, approval status, and attached verification evidence. The Chat Tasks panel keeps full plan progress visible, lets tests/logs/screenshots/PRs/releases be attached as task evidence, and includes a collapsible past-plan archive for the active session. The Chat Git Inspector panel detects the active session repository and shows branch, remotes, staged/unstaged files, log, branches, file diffs, and approved mutation controls for stage, unstage, discard, commit, and branch switch actions without leaving the console. The Chat Skill Inbox extracts reusable skill candidates from the active transcript, shows message-range evidence, and turns approved candidates into local skill drafts. Session Config includes a permission change preview for tool, group, skill, and MCP overrides, automation consent controls for stalled-chat auto-resume delay, allowed resume modes, approved git mutations, and autonomous workspace mutations, plus style sliders for directness, humor, caution, and consent-bounded autonomy. The global Plans page lists active plans across sessions, while global archive data is available through /v1/admin/plans/archive for future planning views.

Group	Page	Path	Purpose
Home	Mission Control	/console	Live overview for Pulse, Reflection, active plans, Agent Runtime runs, Cron jobs, disk pressure, active sessions, recent notifications, recommended setup actions, and release/PR shortcuts
Work	Chat	/console/chat	Interactive agent chat with first-turn cost/quality tier recommendation, tool calling, dockable Sessions and tool panels, session health recommendations, task contract review/approval, task evidence attachments, Git Inspector with approved mutation queueing, Skill Inbox extraction from the active transcript, message-level session forking, @ file/directory/subagent mentions, / command popover for client commands and skills, transcript-snippet session search, parallel/compare subagent progress cards, Files workspace shell, Prior Context preview, and advanced /config session policy, style, and automation consent overrides
Work	Lineage	/console/sessions/graph	Git-log-style session tree showing root sessions, forked children, fork point previews, direct navigation back into the selected chat, and fork insight promotion into Memory Inbox
Work	Plans	/console/tasks	Review active plans across sessions with progress cards and jump directly into the owning chat session
Work	Memory	/console/memory	Review extracted memory candidates before storage, edit stored knowledge assets with inline guidance, inspect fill/read metadata, and compare Tool path vs Prefetch path recall
Work	System Prompt	/console/sysprompt	Edit USER.md, IDENTITY.md, AGENTS.md, TOOLS.md with starter templates, prompt impact metadata, preview, and a technical details toggle
Work	Extensions	/console/extensions	Skills, hub quality score/trust signals, sandbox-tested hub installs for skills/plugins/MCP packages, local Skill Creator drafts/tests, local MCP Server Creator drafts/tests, plugins, MCP servers
Operate	Agent Runtime	/console/agentruntime	Inspect subagent run history with filters, list/tree/Gantt/Flow views, originating session links, cost summaries, replay scrubber, per-run cost/token flow, file attention, git diff timeline, checkpoint restart controls, compare-mode run links, run-derived profile recommendations, and subagent tier management
Operate	Approvals	/console/approvals	Review risky cleanup plans and approved Git mutations before TARS applies them, then inspect the Automation Audit log
Operate	Cron	/console/cron	Manage global scheduled jobs with delivery targets, pause/resume, run-now, delete, and run history
Operate	Logs	/console/logs	Tail configured runtime logs with file, level, component, line-count, refresh, and auto-refresh controls
Operate	Analytics	/console/analytics	Visualize usage totals, daily token bars, model cost rows, and tool or skill call counts
Operate	Pulse	/console/pulse	Watchdog status, incident cards, and run-now trigger
Operate	Reflection	/console/reflection	Nightly batch status and run-now trigger
Setup	Settings	/console/config	Quick Start onboarding plus structured object/array editing, typed LLM tier editing, subsystem-aware pending-change impact previews, and field metadata badges

Settings field metadata, subsystem impact hints, preferred YAML patch paths, and compatible nested YAML parsing are kept together in the config field registry so console schema and config file updates stay aligned. Pending changes also get a frontend fallback classifier so new fields still show the affected subsystem before save.

Requirements

• Go 1.25.6+ (for building from source)
• LLM provider credentials (Anthropic, OpenAI, Gemini, or Claude Code CLI)
• Optional: Gemini API key for semantic memory embeddings
• Optional: Node.js for Playwright browser automation

Build

make build-bins
bin/tars version

For development with hot-reload:

make dev-console    # Vite (5173) + Go API (43180), open http://127.0.0.1:43180/console

Documentation

• Project homepage
• Getting Started
• Plugin and MCP Packaging Guide
• Contributing
• Changelog

Status

Pre-1.0.0 — Module path: github.com/devlikebear/tars