briar-cli 1.1.49

Terminal client for the Briar agent-orchestration API.

briar

Turn the live state of your tools into agent-ready context — then let autonomous agents act on it. All on your machine.

briar is a local-first Python CLI that mines what's actually happening across your stack — GitHub, Bitbucket, AWS/GCP/Azure, Jira, Linear, Fireflies — into a knowledge store, keeps it fresh on a schedule, and runs LLM agents that fix PRs and ship tickets against it.

No SaaS, no remote workspace, no data leaving your laptop. Your credentials, your machine, your APIs — briar just shells out to them directly and writes the results to local markdown or Postgres.

pip install briar-cli
briar version

---

Why briar

• 🔒 Local-first. Everything runs on your machine against your own API tokens. Nothing is uploaded to a service.
• 🔌 One CLI, every system. GitHub · Bitbucket · AWS · GCP · Azure · Jira · Linear · Fireflies — behind a single, consistent interface.
• 🧠 Context, not dashboards. Extraction produces clean markdown knowledge an LLM (or a human) can actually read — PR archaeology, codebase conventions, infra inventory, reviewer profiles, meeting digests.
• ⏰ Cron, replaced. An in-process scheduler keeps per-company knowledge fresh with one long-running command.
• 🤖 Agents that do the work. Point an agent at a PR to address review comments, or at a ticket to implement it end-to-end — branch, code, open a draft PR.
• 🗺️ Plans from your board. Turn a Jira/GitHub Projects board into an ordered, LLM-synthesised implementation plan and run it card by card.

---

Quickstart

pip install briar-cli

# 1. Authenticate the providers you'll use (tokens land in ~/.config/briar/secrets.env)
briar auth login github-pat --company acme
briar auth login jira-token --company acme
export ANTHROPIC_API_KEY=sk-ant-...        # LLM key comes from the environment

# 2. Mine a repo's PR history into a knowledge blob
briar extract --company acme \
    --include pr-archaeology \
    --repo acme-co/acme-app --max 50

# 3. Read it back
briar context get knowledge:acme

Telemetry: briar ships with opt-out error/usage analytics (Sentry). No prompts, file contents, ticket keys, repo names, paths, or secret values ever leave the machine. Turn it off any time with briar telemetry off, BRIAR_TELEMETRY=off, or DO_NOT_TRACK=1.

Less typing: shared flags, project config, inference

briar resolves every flag through one chain — CLI flag > env var > project config > built-in default — so the stable values move off the command line:

# .briar.toml (or [tool.briar] in pyproject.toml), searched upward from cwd
company = "acme"
store   = "postgres"

[repo]
owner = "acme-co"
repo  = "acme-app"

With that file present, and inside the git checkout, the same extract is just:

briar extract --include pr-archaeology        # company + repo come from config/git

• Canonical extractor flags. One shared knob per concept — --repo, --since-days, --max, --top-n, --sample, --authors-allow/-block, --assignees-allow/-block — applies to every extractor selected with --include. The old per-extractor flags (--pr-repo, --risk-since-days, …) still work but are hidden from -h; run briar extract --advanced-help to see them.
• One --repo owner/repo everywhere. extract, agent, and plan run all take the owner/repo slug (or a bare name with --owner). Both are read from the git origin remote when neither the flag nor config supplies them.
• Derived defaults. --store defaults to postgres when BRIAR_DATABASE_URL is set, else file (so a laptop with no DB just works). agent implement's --ticket-project is derived from the ticket key for Jira/Linear (ACME-42 → ACME) or from owner/repo for GitHub/Bitbucket Issues.
• Consistent flag names. The knowledge-store root is --root everywhere (the old --knowledge still works); the credential store is --cred-store on auth/secrets (the old --store still works) so it never clashes with the knowledge-store --store. Deprecated spellings print a one-line note.
• A per-extractor override always wins over the shared flag when both are given.

Helpers for the config + setup loop:

briar init                 # write a starter .briar.toml (repo inferred from git)
briar config show          # see each setting's resolved value AND its source
briar doctor               # check config, git, credentials, store (CI-usable exit code)
eval "$(briar completion bash)"   # tab-completion (also: zsh)

Other niceties: briar --version, quiet-by-default logs (logs to stderr, --verbose for DEBUG), and a once-a-day "new version available" nudge (opt out with BRIAR_NO_UPDATE_CHECK=1 / DO_NOT_TRACK=1).

---

What you can do

briar extract — mine live state into knowledge

# PRs + AWS infra in one shot, filtered to your team
briar extract --company acme \
    --include pr-archaeology --include aws-infra \
    --repo acme-co/acme-app \
    --authors-allow alice --authors-allow bob \
    --aws-extract-region us-east-1 \
    --aws-extract-service ecs --aws-extract-service rds

# Account-wide inventory: every tagged AWS resource across all services
briar extract --company acme --include aws-infra \
    --aws-extract-service tagging-inventory

# Last 14 days of Fireflies meeting summaries for an attendee list
FIREFLIES_ACME_API_KEY=ff_xxx briar extract --company acme \
    --include meeting-digest --meeting-since-days 14 \
    --meeting-attendee-allow [e-mail]

# Code-quality signal from git history + the repo-host API.
# One --repo feeds every selected extractor.
briar extract --company acme --repo acme-co/acme-app \
    --include defect-hotspots --include pr-hygiene \
    --include review-nits --include ci-health

Code-quality extractors (all --provider github|bitbucket): defect-hotspots (churn × bug-fix × size risk), pr-hygiene (size/rubber-stamp/time-to-review), review-nits (recurring reviewer asks → lint candidates), revert-signals, commit-message-quality, stale-prs, ci-health, dependency-health, code-scanning, repo-governance, test-discipline, release-cadence, todo-density. See agents/extract.md.

Feed the knowledge to Claude Code — on demand

# Merge a knowledge index into CLAUDE.md; full detail lands in
# .briar/knowledge/<company>.md for the agent to read when relevant.
briar extract --company acme --repo acme-co/acme-app \
    --include defect-hotspots --include ci-health \
    --merge-claude-md

--merge-claude-md writes the full bundle to .briar/knowledge/<company>.md and splices a short, marker-bounded index — section titles plus a pointer to that file — into CLAUDE.md (override with --claude-md-path). Because CLAUDE.md is auto-loaded into every Claude Code session but the detail file is not, the knowledge stays available on demand without paying a per-session context cost: the agent reads the detail file only when a task touches one of the listed topics. Re-runs replace just briar's block, leaving your hand-written CLAUDE.md untouched.

briar runbook serve — scheduled extraction, in-process

Describe every company + task in one YAML and let briar run the schedule forever — no cron, no external job runner.

briar runbook serve runbooks/

briar agent — autonomous LLM flows

# prfix: read a PR's open review comments, push fixes, reply inline.
# --repo takes the owner/repo slug; both are inferred from git in a checkout.
briar agent prfix \
    --company acme --repo acme-co/acme-app \
    --pr 42 --branch fix-typo \
    --runbook runbooks/acme.yaml

# implement: take a ticket end-to-end — clone, branch, code, open a draft PR.
# --ticket-project is derived from the ticket key for Jira/Linear (ACME-42 → ACME).
briar agent implement \
    --company acme --repo acme-co/acme-app \
    --ticket-key ACME-42 --tracker jira \
    --runbook runbooks/acme.yaml

# Preview the exact prompt + tools without spending a token
briar agent prfix --company acme --repo acme-co/acme-app \
    --pr 42 --branch fix-typo --dry-run

briar plan — LLM-driven implementation plans from a board

# Build an ordered plan from a GitHub Projects board, with company knowledge spliced in
briar plan build https://github.com/orgs/acme/projects/1 \
    --name acme-q3 --company acme --llm anthropic --with-knowledge

# Run the loop: the selector picks the next card, the engineer agent ships it,
# the knowledge store learns what changed — card by card.
briar plan run acme-q3 \
    --company acme --repo acme-co/acme-app \
    --tracker github-issues --llm anthropic

# Smoke one card with --dry-run before letting the loop go wide
briar plan run acme-q3 --limit 1 --dry-run --llm anthropic \
    --company acme --repo acme-co/acme-app

briar scaffold — JSON config bundles for downstream tools

briar scaffold implementation \
    --prefix acme-impl --source github \
    --owner acme --repo widgets

Plus briar context (local knowledge blobs), briar dashboard (read-only HTML status page), briar secrets doctor (credential coverage), and briar journal (decision-journal inspection).

Repeatable flags — fan out across repos, projects, services

Most list-style flags accept multiple occurrences: repeat the flag, once per value (there's no comma form — --repo a,b is one repo named a,b). The canonical --repo feeds every extractor selected with --include.

# Mine several repos, keep the team's PRs, drop the bots — one --repo,
# every selected extractor. --max applies per repo; author allow/block
# compose as allow ∩ ¬block.
briar extract --company acme \
    --include pr-archaeology --include defect-hotspots \
    --repo acme-co/web --repo acme-co/api --repo acme-co/mobile \
    --max 75 \
    --authors-block "dependabot[bot]" --authors-block "renovate[bot]"

# Scaffold a triage flow from two sources — one shared author/assignee
# filter applies to every --source.
briar scaffold implementation --prefix acme-triage \
    --source github --source jira \
    --owner acme-co --repo acme-app --github-secret-id <uuid> \
    --authors-block "dependabot[bot]" \
    --assignees-allow alice --assignees-allow bob \
    --jira-project ACME --jira-project PLAT --jira-secret-id <uuid> \
    --auth-mode pat

The same lists map onto runbook YAML as arrays — e.g. repo: [acme-co/web, acme-co/api] under an extractor's args:.

Divergent identifiers in one run. Some extractors key off a tracker project (active-tickets, ticket-archaeology) rather than an owner/repo slug. When you run those alongside repo-based extractors in a single invocation and they need different values, reach for the per-extractor override flags (--ticket-project, --pr-repo, … — listed by briar extract --advanced-help), or split into two invocations. The runbook YAML models this cleanly: one args: block per extractor.

Handy patterns

# Script briar: --format json pipes straight into jq (works on every command).
briar plan status acme-q3 --format json | jq -r '.blocked[].key'

# Gate CI on credential coverage — exits non-zero if any runbook is missing creds.
briar secrets doctor --examples runbooks/

# Cost-safe agent rollout: preview for free, then one paid card, then go wide.
# --ticket-project is derived from the key (ACME-42 → ACME); --repo takes the slug.
briar agent implement --company acme --repo acme-co/acme-app \
    --ticket-key ACME-42 --dry-run
briar plan run acme-q3 --company acme --repo acme-co/acme-app \
    --tracker jira --llm anthropic --limit 1 --max-iter 20

---

Install options

pip install briar-cli                # base: GitHub/Bitbucket/AWS, Jira/Linear, Anthropic + Bedrock, file + Postgres stores
pip install 'briar-cli[openai]'      # OpenAI LLM
pip install 'briar-cli[gemini]'      # Google Gemini LLM
pip install 'briar-cli[mcp]'         # MCP-server tools for `briar agent` (runbook `mcp:` block)
pip install 'briar-cli[gcp]'         # GCP cloud provider
pip install 'briar-cli[azure]'       # Azure cloud provider
pip install 'briar-cli[vault]'       # HashiCorp Vault credential store
pip install 'briar-cli[all]'         # everything

Each adapter fails loudly with the right install command if its SDK is missing. Python 3.10+ (tested through 3.12).

---

Documentation

Full command reference, every flag, runbook-YAML schema, configuration, and recipes:

📖 usebriar.com/docs

• End-to-end usage flows (14 multi-feature recipes — onboard a company, extract AWS + Fireflies + fix a PR, build & run a plan, full lifecycle in one sitting, …): agents/flows.md
• Per-command operator manual: agents/
• A comprehensive multi-company runbook lives in examples/all_features.yaml.

---

License

See LICENSE.