Sidekick
A local-first terminal agent for developers — chat, code, and 27 tools, on your hardware.
No cloud account required. No API bill by default. Your files, memory, and voice stay on your machine unless you hand it a key.
uv tool install "sidekick-agent[voice]" && sk init && sk
Quickstart
uv tool install "sidekick-agent[voice]" # global `sk`, local STT included
sk init # hardware → model → verify → web search
sk # fullscreen chat
Requires Python 3.12+. The local path needs Ollama (ollama serve). sk init
checks your hardware, suggests a model that fits, and verifies the connection.
Any OpenAI-compatible endpoint also works — bring a key or point Sidekick at a
local server.
See it
Review the diff before you commit. sk review and /review run one
evidence-backed engine: every finding needs a real file:line and an exact
evidence quote, or it is dropped. Ten built-in rules cover common security and
correctness footguns, each with a positive/negative fixture; the eight
security/* rules are protected from project overrides. Branch and PR reviews
also get an advisory worth-it assessment, and sk review --verify runs syntax
plus the project's configured checks with a separate verdict. /diff prints the
same diff without a model call. See Review.
Approve the plan, then verify the result. Sidekick reads the repo, asks you
to approve the multi-tool plan, writes the test, runs it, and reports the
result. After edits it also runs syntax checks plus the project's trusted
lint/typecheck checks, with timeouts and missing tools reported as unknown —
never pass. Reads run automatically; writes and shell commands ask first.
Real headless captures of the shipped TUI from docs/shot.py.
The demo sessions are seeded and no model is called; qwen3-coder:30b is a
display model, and sk init picks one for your hardware. The marketing/docs
site lives at https://faisal-fayaz.github.io/sidekick/ (source in
website/).
Why Sidekick
Sidekick is a terminal agent that can actually operate your machine: read and edit code, run commands, search the web only where you allow it, remember projects, and leave an audit trail you can inspect.
It is built for the cases where a hosted agent is a bad fit: private repos, regulated code, offline or flaky networks, small GPUs, and developers who want to see exactly what the agent did.
| Sidekick | Typical hosted agent | |
|---|---|---|
| Runs fully offline with Ollama | ✅ | ❌ |
| Source code stays on your machine by default | ✅ | usually sent to a provider |
| Voice transcribed on your CPU | ✅ | ❌ |
| Answers grounded in your system, not guessed | ✅ deterministic grounding | prompt-only |
| Approval gates, hard refusals, and checkpoints | ✅ | varies |
| Deny-by-default egress for model-initiated fetches | ✅ sk egress, logged |
❌ |
| Audit ledger with credentials redacted | ✅ sk audit |
vendor-side logs |
| MCP server and client | ✅ both directions | varies |
| Editable skills, commands, and tools | ✅ SKILL.md, TOOLS.md |
varies |
| Default bill | $0 | metered |
Developer workflow
Sidekick is deliberately approval-first. Reads run automatically; writes, deletes, and general shell commands ask. That makes it useful on a real repo without handing an unattended model the keys.
| Task | What you use |
|---|---|
| Explore a repo | list_dir, repo_map, read_file (with line ranges), search_code, glob_files, and the read-only exec allowlist |
| Edit safely | write_file / edit_file with approval; /plan to propose; /rewind to undo |
| Inspect changes | /diff, git_status / git_diff / git_branch, and sk review / /review with evidence-backed findings |
| Verify changes | Automatic post-edit syntax + lint/typecheck verification, sk check / sk check --changed for the full suite, and sk review --verify for a diff |
| Automate | sk run --json, sk run --bg, and sk jobs |
| Extend | skills, custom slash commands, declarative TOOLS.md tools, MCP |
| Stay in control | approval cards, hard refusals, egress allowlist, a declarative sandbox (plus bubblewrap/Seatbelt when installed) for what a command may reach, audit ledger |
The TUI and plain-text REPL share every command, so a scripted fallback and an interactive session behave the same way.
The git tools take typed arguments, not a command string: git_status,
git_diff, and git_branch are approval-free reads; git_commit and
git_push ask, like any other write, and git_push refuses a protected
branch. Because the command line is built by the tool, there is nothing for a
model to smuggle into it.
search_code uses rg when installed, then grep, then a Python walker. It
skips hidden paths, symlinks, binaries, and .gitignore matches; results default
to 50 (hard cap 200) and say when they are truncated. glob_files returns at
most 200 sorted paths. repo_map outlines a directory tree with language
totals, file sizes, and top-level symbols; its default budget is 1,500 tokens
and it reports token, node, and scan truncation. None of these needs approval.
Chat
One input, two surfaces:
sk # fullscreen TUI with streaming + themes — start here
sk tui --model fast # same, explicit form
sk chat # fallback REPL: dumb terminals, screen readers, broken TUIs
Type / and autocomplete filters all 30+ commands — Enter completes, Tab too,
Esc dismisses, ↑/↓ navigates. F1 opens a generated cheatsheet built from the
same tables as the dispatcher, so it cannot rot.
TUI keys: Enter sends · ctrl+j/alt+enter newline · ↑/↓ history · ctrl+y copies · ctrl+g push-to-talk · pgup/pgdn scroll · F1 help · F2 dark/light theme · F3 sessions drawer · F4 plan/build toggle · F5 model picker.
Answers stream live as Markdown. The status bar shows model, session, mode, and last-turn stats. Approvals arrive as cards with a timeout.
Plan mode proposes without writing: /plan blocks file writes in chat
(/build reverts), F4 toggles the same in the TUI, and sk run --plan does it
single-shot. Sessions carry /compact [hint], /diff, /review, and
/rewind [n] to undo agent file edits. sk review --pr N can post comments to
GitHub or Bitbucket when credentials are available; it never edits the branch.
Install
| Channel | Command |
|---|---|
| PyPI / uv | uv tool install "sidekick-agent[voice]" |
| PyPI / pipx | pipx install "sidekick-agent[voice]" |
| PyPI / pip | pip install "sidekick-agent[voice]" |
| AUR (Arch) | yay -S python-sidekick-agent |
| conda-forge | conda install -c conda-forge sidekick-agent (separate feedstock repo) |
The published name is sidekick-agent; the command stays sk. Without
[voice] you get everything except Talk/mic, which installs on first use.
Version lives in src/sk/__init__.py. Publishing is automatic and
credential-free: merging a version bump to main trusted-publishes to PyPI and
opens a GitHub Release. Forks cannot publish. Details:
packaging/README.md.
From source (dev):
git clone https://github.com/Faisal-Fayaz/sidekick && cd sidekick
uv tool install -e ".[voice]"
sk doctor
Voice
sk talk [-d SECS] [--stt-model base] [--device hw:2,0] # Enter records, Enter stops
sk mic-test # peak dB + verdict
Capture uses the OS-native recorder (arecord/ALSA on Linux, sox/ffmpeg on
macOS); transcription uses local faster-whisper int8. The transcript lands
editable in the prompt. In the TUI, ctrl+g or the mic pill does the same.
Recordings are temp files and are deleted after each take.
MCP
sk mcp [--allow-writes] # JSON-RPC 2.0 over stdio, zero new deps
All 27 tools, same safety policy: SSRF guards, write blocklists, and hard
refusals apply unchanged. Reads auto-run; shell/writes/delete need
--allow-writes, otherwise they return a clean denied error. Stdout carries
protocol only.
Claude Desktop snippet:
{ "mcpServers": { "sidekick": { "command": "sk", "args": ["mcp"] } } }
Sidekick also works as an MCP client. See docs/mcp-client.md.
Providers
sk connect # pick provider → paste key (hidden) → pick model → ping
Presets: ollama | openai | groq | together | deepseek | openrouter | google | lmstudio | anthropic | opencode | custom. anthropic speaks the native
Messages API; the rest are OpenAI-compatible. Any compatible endpoint works via
--provider custom --base-url https://... --api-key ....
Preferred key handling: SIDEKICK_API_KEY in the environment (never touches
disk). File keys are chmod 600 and masked in --show. sk auth add/list/status/remove manages credentials; sk model asks the provider for
its live list.
Reasoning effort: sk config --reasoning-effort low (default;
off|minimal|low|medium|high|max), escalated to high in plan mode. Honored on
OpenRouter (reasoning.effort), OpenAI (reasoning_effort), Anthropic
(thinking budget sized to fit max_tokens), and Ollama (think toggle).
OpenCode Zen free tier
The opencode preset points at OpenCode Zen, with free tool-capable models plus
paid tiers. The anonymous free tier is restricted to opencode's own app, so add
a free OpenCode account key first: OPENCODE_API_KEY or
sk auth add opencode (get it at opencode.ai/auth). Fast/smart resolve to
nemotron-3.5-lightning-free / muse-spark-1.3-contributor-free; run
sk models opencode for the live list.
Command reference
Chat and sessions
| Command | What |
|---|---|
sk / sk tui [--continue] [--model M] [--allow LIST] [--deny LIST] |
Fullscreen chat, fresh session each launch. sk --version prints the version and exits; --cwd DIR and --profile NAME work on every command |
sk chat [--continue] [--session S] [-y|--yes] [--model M] [--allow LIST] [--deny LIST] [--no-stream] |
Fallback plain-text REPL (dumb terminals, screen readers, TUI issues) |
sk talk [-d SECS] [--stt-model base] [--device hw:2,0] [--session S] [-y] [--install] |
Push-to-talk voice chat (CPU transcription; --install skips the install prompt) |
sk mic-test [-d SECS] [--device hw:2,0] |
Mic level check: peak dB + silent/quiet/good verdict |
/sessions, /resume <n>, /sessions delete <n>, /fork [n] |
List, switch, delete, branch past sessions |
/yolo, /confirm, /readonly |
Approval modes: auto-approve writes / ask every time / block all file writes |
/compact [hint], /diff, /review |
Summarize history on demand / inspect working-tree diff / evidence-backed review |
/plan, /build |
Plan mode: propose without writing / back to build mode |
/rewind [n] |
Undo an agent file edit (snapshots write/edit/delete targets; shell mutations are not tracked) |
Agent and automation
| Command | What |
|---|---|
sk run "task" [--yes] [--plan] [--read-only] [--deny LIST] [--model auto|fast|smart|name] [--json] [--bg] [--allow LIST] [--session S] [--no-stream] [--trust-project-checks] [--verify-full] |
Single-shot agent run. Auto-routes the model; --json emits one machine-readable document + exit codes; --bg detaches and notifies; --allow shell:pytest,write_file skips prompts for listed tools. After edits it runs syntax checks plus trusted lint/typecheck checks; --verify-full runs every configured check |
sk review [--base REF|--staged|--commit SHA|--pr N] [--path P] [--intent TEXT] [--assess|--no-assess] [--post|--no-post] [--json|--sarif] [--fail-on LEVEL] [--explain|--check-config] [--trust-project-rules] [--verify [--yes|--allow LIST|--deny LIST]] [--disable RULE] [--severity RULE=LEVEL] |
Evidence-backed review of a diff or PR. Findings need file:line evidence; the worth-it assessment is advisory. Policy comes from layered review packs (~/.sidekick/review.toml, then a trusted repo pack). --pr auto-posts comments to GitHub (via gh) or Bitbucket when credentials exist; it never edits the branch. --verify adds syntax + configured lint/typecheck verification for the reviewed files and exits 1 on fail/unknown |
sk fix "<problem>" [--json] [--yes] [--allow LIST] [--deny LIST] [--trust-project-checks] [--verify-full] [--session S] |
Investigate and fix a problem, then prove it: the agent works under the approval gate, its edits are checkpointed, and the changed files are verified (syntax + the project's lint/typecheck checks). Reports the verdict — pass, fail, or unknown — and exits 1 when the change did not verify |
| `sk serve [--host 127.0.0.1] [--port 8787] [--read-only] | A localhost HTTP surface over the same agent loop, tools, and audit trail. Loopback only — never 0.0.0.0 — and always behind a token |
| `sk explain "" [--json] [--session S] | Explain the code, grounded in it: the agent reads the repository and answers with file:line citations, saying plainly when the code does not answer the question. Read-only — no write and no shell is reachable from it |
| `sk pr [--base REF] [--title T] [--body B] [--remote NAME] [--push] [--yes] [--json] | Prepare a pull request from the working tree. The title and body are derived from the branch's commits and the diff — they cannot claim work that did not happen. With --push the branch is pushed and the pull request opened through gh, both after a confirmation; protected branches are never pushed |
sk test [--command CMD] [--json] [--yes] [--allow LIST] [--deny LIST] [--trust-project-checks] [--max-diagnostics N] [--session S] |
Run the project's tests through the same check pipeline: the declared test check, or a detected one (Makefile, package.json, cargo test, go test ./..., pytest -q). Same approval and audit path; a missing runner is unknown, never a pass |
sk check [--changed] [--json] [--yes] [--allow LIST] [--deny LIST] [--trust-project-checks] [--check-config|--explain] [--max-diagnostics N] [--session S] |
Run .sidekick/checks.toml through the normal approval and audit path. Parses pytest, ruff, mypy, eslint, tsc, and go test into compact diagnostics; unparsed output stays raw. Exits 1 on failure or unknown (timeout, missing tool, denial), 2 on config/trust errors. Untrusted by default. See Checks |
sk jobs [-n N] |
List background jobs from sk run --bg |
sk brief [-p PATH] [--smart] |
Morning digest: system + git + todos + memories, instant without an LLM |
sk digest [--force] |
Brief + overnight failures, desktop nudge or log |
sk export [SESSION] [--out f.md] [--force] |
Session transcript as Markdown. Refuses to overwrite unless --force |
sk imagine "prompt" [--out f.png] [--size WxH] [--model M] |
Generate an image via the provider images endpoint |
Memory and search
| Command | What |
|---|---|
sk remember/recall/memories/forget |
Long-term memory (FTS5 search, auto-injected) |
sk search QUERY [--session S] |
Full-text search across past transcripts |
sk todo add/list/done/clear |
Todos |
sk history / sk oops |
Shell log / explain last failure |
sk stats [--session S] [--format md|json] |
Usage + cost estimates from audit rows; pair with spend caps |
Safety and audit
| Command | What |
|---|---|
sk egress [list|allow HOST|deny HOST|test URL] |
Egress policy: what the model may fetch, and what it was blocked from (docs) |
sk audit [--session S] [--format md|json] [-n N] |
Compliance log: tool runs, approve/deny, local-vs-egress (-n clamped 1–1000) |
sk hook-install [--write] |
Bash/zsh logging hook |
sk hooks [--check] |
List event hooks + live dry-run (see docs/hooks.md) |
Extensions and servers
| Command | What |
|---|---|
sk skills / sk skills-search / sk skills-registry [QUERY] / sk skills-install NAME |
Skill packs: registry + superpowers |
sk plugins |
User-defined tools (TOOLS.md, see docs/plugins.md) |
sk mcp [--allow-writes] |
MCP server over stdio (27 tools, safe defaults) |
sk mcp-servers |
List configured MCP client servers + live tool check (see docs/mcp-client.md) |
sk daemon [--once] / sk daemon-install [--schedule TXT] / sk daemon-install-macos [--schedule TXT] / sk daemon-uninstall / sk daemon-schedule [--set TXT] |
Background watcher (systemd/launchd, calendar schedules). It watches the repositories and says something when they change — the branch, the files, how far it is from its upstream — and stays quiet when nothing has |
Setup and system
| Command | What |
|---|---|
sk init / sk setup / sk connect |
Guided first-run / full setup / provider key flow |
sk doctor [--fix] / sk report / sk upgrade [--check] |
Health (+auto-remediation) / redacted diagnostics bundle / self-update |
sk models [list | pull <id> | prune <id>] / sk model / sk config / sk version |
Models (list, download, remove) / ask the provider for its live list / settings / build |
sk auth add/list/status/remove |
Manage provider keys (masked, validated live) |
Packs use the SKILL.md frontmatter format. The prompt carries a relevance-ranked
index; the agent loads full instructions on demand via the skill tool.
fast/smart resolve per provider.
The --json envelope
sk run --json prints exactly one JSON document on stdout, and the exit code
reports the outcome:
{"ok": true, "answer": "...", "model": "qwen3-coder:30b", "session": "default", "tools": ["read_file"], "error": null, "error_code": null}
ok is the outcome of the turn, not of the command. A turn that died
mid-flight used to hand back its error text as the answer with ok: true, so a
caller could not tell a finished run from a throttled one — a benchmark harness
scored a rate-limited run as a completed turn. error_code is a stable
classification so a caller never has to parse prose: rate_limit, auth,
server_error, timeout, network, bad_request, turn_failed, job_failed.
Exit code 1 means the turn failed, 0 that it completed.
Architecture
flowchart TB
U([you]) --> CLI[sk / sk run]
U --> TUI[sk tui: autocomplete, streaming, mic pill]
U --> VOICE[sk talk: arecord + faster-whisper]
CLI --> SLASH[slash.py: /commands, no LLM]
TUI --> SLASH
VOICE --> AGENT
CLI --> AGENT[agent.py: stream → tools → synthesize]
TUI --> AGENT
AGENT --> GROUND[deterministic grounding: ~/paths, URLs,\nsysinfo — injected before the model sees the prompt]
AGENT --> TOOLS[tools/: 27 tools, allowlists,\nhard-blocks, SSRF guard]
AGENT --> MEM[(store.py: history, memories FTS5,\ntodos, shell log)]
AGENT --> SKILLS[skills: relevance-ranked SKILL.md index]
Design bets that paid off: deterministic grounding beats prompt instructions (small models ignore rules but cannot argue with injected facts), text-JSON fallback (coders emit tools as text over the OpenAI endpoint), FTS5 over vectors (zero deps, instant, no embedding server on a 4GB box), parallel reads (approval-gated tools stay serial; independent reads run concurrently with failures isolated), capability profiles over hope (known models declare their tool protocol; the router reads them instead of paying a probing 400).
Safety
Approvals first. Reads auto-run. Writes, deletes, and general shell need
approval (inline card in the TUI, prompt in the CLI), HOME//tmp only, ≤100KB,
never ~/.ssh, ~/.gnupg, /etc, /usr. Multi-tool turns with destructive
actions get one plan review up front instead of per-tool prompts. shell
hard-refuses rm -rf /, mkfs, dd to devices, and fork bombs even with
approval. read_url/web_search block localhost/private IPs.
Egress is deny-by-default. Sidekick will not fetch a destination the model
asked for until you allow that host. sk init asks once whether to enable web
search (html.duckduckgo.com — one fixed host Sidekick picks, not the model);
everything else stays blocked:
sk egress # what is allowed, and what was blocked
sk egress allow example.com # allow one host
sk egress allow '*.example.com' # or a whole domain
sk egress deny example.com # revoke
sk egress test https://example.com/x # check a URL without fetching it
This covers every destination the model can cause a fetch of: read_url,
web_search, and provider-supplied image URLs. With an empty allowlist, all
three are refused. The allowlist is checked on the initial URL and on every
redirect hop, so an allowlisted host cannot redirect the agent somewhere
unlisted. web_search itself only requests html.duckduckgo.com and returns
result URLs as text — it never fetches a hit, so following one is read_url,
which is governed on its own.
Your model API endpoint and the MCP servers you configured are unaffected — you
chose those, they are not something a prompt can talk the agent into adding.
Fetch decisions are written to the ledger with the host and reason, visible via
sk egress and sk audit.
The ceiling, stated plainly: this governs fetches made through the agent's
tools. A shell command running curl is not covered by the egress list — so
the sandbox covers it instead: with a
~/.sidekick/sandbox.toml in place, network access from the shell is denied
unless the host is on the allow list. What a program does after it starts is
still beyond a policy that reads command text, which is why
issue #329 (OS-level
isolation) is the real answer rather than another pattern list. See
docs/egress.md.
Your data is private to your account. ~/.sidekick is 0700 and
everything inside it is 0600 — transcript, shell commands, checkpoints, and the
last 100 prompts. An existing install is tightened the first time you run any
sk command. This is a local filesystem control: it does not protect against
root, and it does not stop anything the agent itself can read.
Benchmarks
Measured, not folklore. A fixed task set — a greeting, a short-answer instruction, a grounding one-liner, a tool turn — run against real models on a 4 GB box (Ryzen 7 4800H, 7 GB RAM, GTX 1650), 2026-11-12, Sidekick 0.33.0.
| target | pass@1 | wall/task | tokens | cost |
|---|---|---|---|---|
ollama/llama3.2:3b |
0.50 | 8.2s | 91 | no API bill |
ollama/qwen2.5-coder:7b |
0.50 | 40.0s | 83 | no API bill |
ollama/qwen3:4b |
0.50 | 87.8s | 174 | no API bill |
A task that needed a tool and never called one is a failure, whatever the
answer says — an ungrounded answer is a guess. That rule is why the two
tool tasks are failures here and not passes: one model answered the grounding
question without calling sysinfo (right by luck, not by measurement), and
another narrated a file write it never performed. A local model has no API bill
— the cost is electricity, which is not on the invoice.
Methodology, per-task detail, and the caveats: docs/eval/results.md.
Regenerate it yourself:
python3 docs/probe-evals.py --compare --repeat 3
Tests
uv run --python 3.12 --with ".[test]" pytest tests -q
The suite covers unit, regression, Textual pilot, property, and security tests
with no Ollama needed. The eval harness (tests/test_eval.py) locks past quality
bugs into offline regression tests, and a suite-wide fixture guarantees tests
never touch your live ~/.sidekick/.
CI runs the suite on Python 3.12, 3.13, and 3.14 across Ubuntu and macOS, plus lint, type-check, build, coverage, and website checks.
Config
~/.sidekick/config.toml holds provider, model, base_url, api_key, and
the rest. Data stays home: history.db, skills/, nudges.log,
input_history, tui-errors.log.
| Env var | Effect |
|---|---|
SIDEKICK_PROVIDER / SIDEKICK_MODEL |
override the configured provider/model |
SIDEKICK_BASE_URL / SIDEKICK_API_KEY |
override the endpoint / key for this process |
SIDEKICK_SPEND_CAP |
per-session USD cap, 0 = unlimited |
SIDEKICK_REASONING_EFFORT |
reasoning effort for backends that support it |
SIDEKICK_VERIFY_REPAIRS |
post-edit verify repair rounds (default 2, 0 disables) |
SIDEKICK_VERIFY_FULL=1 |
after edits, run every configured check instead of syntax + lint/typecheck |
SIDEKICK_PROFILE |
load a named config profile |
OPENCODE_API_KEY |
key for the opencode provider |
SIDEKICK_TRUST_REPO=1 |
opt in to repo-supplied slash commands, review rules, and check commands |
SIDEKICK_TRUST_REPO— why your repo's commands, rules, and checks do nothing. A cloned repo can ship.sidekick/commands/*.md, and a!cmd`` expansion inside one reachessh -cwith no approval prompt. Review is the same shape: `.sidekick/review.toml` can change severities, budgets, and rule guidance, and `.sidekick/checks.toml` declares shell commands for `sk check`. All are ignored unless you set `SIDEKICK_TRUST_REPO=1` in your own shell (or pass `--trust-project-rules` to `sk review` / `--trust-project-checks` to `sk check` or `sk run`). Trust only loads the configuration; every check command still goes through approval and the audit ledger. The flag is read from the environment and never from repo config, so a repository cannot self-authorize (#287). Commands in `/.sidekick/commands/` and the review pack in `/.sidekick/review.toml` always load. See Review and Checks.
History budget: the prompt budget bounds the whole assembled prompt — system
prompt, auto-context, tool schemas, and the reply reservation — then history gets
what is left (#308). history_budget_tokens (default 3000) caps that share;
over-budget sessions compact to a rolling summary via the current model (DB
history stays complete). Lower it for small-context models.
Context window: context_window (0 = take it from the model profile) sets
the total window, and the same number is sent to Ollama as num_ctx, so the
budget and the server cannot disagree. Local models default to 16k, frontier to
200k. The fixed cost alone — system prompt plus tool schemas — measures ~3,700
tokens, so a window below ~8k leaves no usable conversation. Raise it if you
have VRAM to spare; the KV cache grows with it. Run /context to see where a
turn's tokens go.
Per-project config: a .sidekick.toml in any repo layers over the global
file (nearest one walking up from cwd). It may carry a [project] table only:
[project]
docs = ["CONTRIBUTING.md", "docs/egress.md"] # injected into the prompt
memory_namespace = "my-repo" # scopes this repo's memory
Nothing at the top level of a project file is honoured — not provider,
model, max_steps, temperature, egress_allow, or anything else. A repo
you just cloned must not be able to steer the agent's model, budget, approvals,
or network. Those keys live in ~/.sidekick/config.toml or the environment
only. The same applies to [project].approved_commands (shell approvals stay
global). Every ignored key is reported rather than dropped quietly.
sk --cwd PATH runs any command as if in that directory.
Project memory auto-discovery: from the cwd upward, the first SIDEKICK.md
/ AGENTS.md / CLAUDE.md / GEMINI.md found (root-down) is injected into the
prompt automatically.
Config profiles: named ~/.sidekick/profiles/<name>.toml files replace
config.toml when selected via sk --profile NAME … or
SIDEKICK_PROFILE=NAME (env still wins, project files still layer). Manage with
sk config --profiles (list) and sk config --save-profile NAME (snapshot
current non-secret settings — keys stay in keyring/env).
Documentation
- Egress policy
- Review
- Checks
- Sandbox
- The SDK and the local API
- Event hooks
- MCP client
- Plugins and declarative tools
Roadmap
See ROADMAP.md — the shared plan (vision, next milestone, done
list). It changes by pull request only. Contributions are welcome; see
CONTRIBUTING.md.
License
MIT — do what you want, shout-outs appreciated.
Metadata
Release files for sidekick-agent 0.37.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| sidekick_agent-0.37.0.tar.gz | 1.5 MB | With provenance | Details for sidekick_agent-0.37.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| sidekick_agent-0.37.0-py3-none-any.whl | Python 3 | none | any | Details for sidekick_agent-0.37.0-py3-none-any.whl |
Total release size: 1.8 MB
Release files / sidekick_agent-0.37.0.tar.gz
| Download URL | sidekick_agent-0.37.0.tar.gz |
|---|---|
| Size | 1.5 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
93506dd234cbd769cd16dc395298913531d76c4e0fd52de10de9d95e74435047
|
|
BLAKE2b-256 checksum How to use checksums |
e0cc3d57de712caa5c16904f4f2537f7667e9ec1fdc50dd3c4c93d4b25ff1b32
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 14, 2026.
Transparency logRelease files / sidekick_agent-0.37.0-py3-none-any.whl
| Download URL | sidekick_agent-0.37.0-py3-none-any.whl |
|---|---|
| Size | 377.3 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
62cc598f2f88c804806001616b7d797aace15e702b079ccebe6ad090238d526a
|
|
BLAKE2b-256 checksum How to use checksums |
2c25b1c0166b27a10e9a55e53b45822bff73f3e83835e21a7b2ebbd0cf7b2dc8
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 14, 2026.
Transparency logSecurity for sidekick-agent 0.37.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
Both files in this release have been uploaded with provenance metadata.
Attestation for sidekick_agent-0.37.0.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 14, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.