Coding agent · Sigmund Freud
Id
Full-capability, metered, bottom-first coding agent. Shell command
Id (capital I — not Unix id). Fuel from
gateway.idlidu.dev. Not a VS Code fork —
CLI + API client + thin IDE embeds.
- Cheap by default — Bottom fishing; ≥85% stay on Tier‑0/1
- Honest burn — prepaid hard stop; named fees; no silent padding
- Full surface — ask / plan / agent / debug, tools, parallel, MCP
- Grounded learning — consented usage, behaviour, thinking, writing traces
Install
Needs Node.js 20+. One-liner puts Id on your PATH.
curl -fsSL https://www.idlidu.dev/install.sh | bash
source ~/.config/Id/shellenv
Id health With a key already minted:
curl -fsSL https://www.idlidu.dev/install.sh | ID_API_KEY=idl_… bash Artifacts: Id-latest.tar.gz · version · install page · source manishthatte/id
API keys
Mint at app.idlidu.dev/keys.
Keys are idl_…, shown once. Each key has:
- Budget ($ / turn) — default Id turn budget (override with
--budgetor/budget) - Hard limit ($ lifetime) — gateway rejects further burns once spent would exceed this
- Spent — running lifetime spend against that hard limit
Account credits are separate (prepaid balance). Hard limit is per key.
Revoke any key from the same page — revoked keys get 401 on chat/agent.
export ID_API_KEY=idl_…
export ID_GATEWAY=https://gateway.idlidu.dev # optional
export ID_API=https://api.idlidu.dev # optional
# or: ~/.config/Id/env (mode 600) Modes
Ask
Read-only tool allowlist — explore without mutating.
Plan
Propose a plan; approve before write / shell mutate.
Agent
Full tools — edit, shell, explore, MCP, browser.
Debug
Instrument → evidence → fix loop.
Permissions: manual · accept_edits · plan ·
auto · yolo. Headless -p does not auto-approve tools.
Commands
CLI
Id # interactive REPL
Id ask "Explain README" --budget 0.05 --yolo
Id plan "Add soft-cap UI" --budget 0.15
Id run "Fix the failing test" --budget 0.25
Id -p "…" --budget 0.10 # headless (does not auto-approve tools) REPL slash
/ask /plan /agent /debug switch mode
/budget 0.25 per-turn prepaid cap
/model bottom or anthropic/… · pin
/permissions manual|auto|yolo
/steer <text> mid-run follow-up
/status /history /copy /help /exit Also
Id models·Id chat·Id health·Id metricsId sessions·Id resume <id>Id --explore "map packages" run "…"·Id best-of-n "…" --best-of-n 3Id skills list|search|install·Id import-rulesId pr-fix [--watch]·Id handoff "…"·Id serve [--acp]Id soft-cap [usd]·Id insights·Id update
Flags: --budget --max-steps --model --effort --parallel --review-diffs --checkpoint --shell-net --no-telemetry …
Bottom & budgets
Default model is Bottom — gateway picks the cheapest capable model.
Pin with --model anthropic/… or /model, or raise
--effort high.
- Turn budget — stops the local loop when the turn spend hits the cap
- Key hard limit — lifetime spend on that
idl_key - Account balance — prepaid credits; hard stop at $0
- Soft monthly cap — warn / stop locally (
Id soft-cap); bypass with--force - Low-balance alert — warn at ≤15% of last top-up (and same % of key hard limit); change in app Settings
More: Bottom routing · Model catalog · Pricing
Tools & safety
- Local FS: read / write / edit / grep / glob ·
.idignore+ gitignore - Sandboxed shell (bwrap; network off by default)
web_fetchSSRF-safe · Playwright MCP browser · mail (Thunderbird Gloda, read-only)- Explore / task subagents under a shared tree budget · best-of-n worktrees
- Rules:
AGENTS.md·.id/rules/· skills · hooks · pending diffs - Config:
.id/permissions.json·.id/modes.json
Telemetry
Primary objective: consented structured traces — usage, behaviour, thinking, writing —
flushed to idlidu.dev. Args are redacted; crash-safe flush. Opt out:
--no-telemetry or env.
Embeds
Id serve— JSON-RPC for VSCodium extensionId serve --acp— ACP-shaped stdio (Zed and friends)- Neovim example in the id repo · JetBrains TBD
Spending
Prepaid packs on the dashboard. See live balance, stay-rate, and burn there — this page stays the product reference.