@timps-ai/timps-code

CLI coding agent with persistent memory across sessions. 47 tools, 6 LLM providers, 22-layer memory, swarm mode, multimodal vision, and a self-improving agent loop.

npm install -g @timps-ai/timps-code
timps

What it does

• 22-layer memory — working, episodic, semantic, procedural + 18 forge layers (Chronos, Echo, HarmonicSheaf, Aether, EngramLog, Consolidation, SynapticPruner, Provenance, SpacedRepetition, ConstitutionalGuard, Audit, ProspectiveTrigger, BiasRevealer, ContextVector, Rehearsal, SchemaDistorter, ConfidenceCalibrator)
• 47 agent tools — file ops (read/write/edit/patch/multi-edit/search/list), git (status/diff/log/commit/stash), shell (bash), web (search/fetch), reasoning (think/plan), MCP, todo, project info, diagnostics, browser automation, workflow pipeline, remote trigger, cron scheduling, team management
• 6 providers — Ollama, OpenAI, Gemini, Claude, OpenRouter, hybrid fallback chain
• Swarm — 10 agent roles for multi-agent DAG execution
• Self-improving agent — tracks mistakes, builds prevention instructions, GRPO training data
• Multimodal memory — image embedding (Gemma 3 via Ollama), audio reference, diagram storage, terminal capture, cross-modal recall
• LSP proxy — wraps real language servers, injects contradiction + bug-pattern diagnostics
• MCP client — auto-discovers 15+ MCP servers (postgres, sqlite, redis, github, slack, sentry, puppeteer, etc.)
• WASM sandbox — install/uninstall plugins with permission enforcement
• LLM key vault — AES-256-GCM encrypted API key storage
• Provider rate limiter — per-provider sliding window, daily cap, automatic fallback
• Keybinding-powered TUI — Ink/React 19 terminal UI with streaming

Install & Run

npm install -g @timps-ai/timps-code
timps                           # Interactive session
timps "add validation to api"   # One-shot
timps --provider ollama         # Specify provider
timps --config                  # Setup wizard

Flags

--provider, --model, --dir, --config, --branch, --merge, --tui, --war-room, --binary-synth

Slash Commands

/help, /memory, /todo, /branch, /merge, /skills, /mcp, /git, /models, /doctor, /clear, /vision, /audio, /screenshot, /diagram, /terminal, /crossmodal, /audit, /improve, /config:encrypt-key, /auth:login, /limits:show, and more.

Architecture

User input → AgentLoop.run() → Agent (plan + execute)
  ├─ 47 tools (file, git, shell, web, MCP, memory, etc.)
  ├─ 22-layer memory (working → episodic → semantic → forged)
  ├─ 6 model providers with fallback
  └─ self-improving loop (mistake tracking → prevention injection)