v0.3.1

Patch release fixing two issues found right after v0.3.0 shipped. If you're on 0.2.x or 0.3.0, this is the version to update to.

Fixed

• burnless update no longer rolls back when your instance is running. The post-update health check counted a busy port as a failure — but during an in-place update your current instance is still serving on that port, so the upgrade would roll itself back every time. Port (and a not-yet-generated secrets key) are now treated as informational, so update completes cleanly whether or not the app is running. burnless start still refuses a port that's truly taken.
• npm package now ships the right build. burnless@0.3.0 was published from a stale build (it reported 0.2.1). The publish path now rebuilds and verifies the version before shipping, so the package on npm always matches its label.

Update

burnless update                                # existing install → 0.3.1
curl -fsSL https://burnless.ai/install | sh    # new install

Verify the download:

shasum -a 256 -c SHASUMS256.txt

Full changes: #16.

v0.3.0

The first feature release since the OSS launch — Burnless now reads your real money. Bundles PRs #11, #12, #13, #14.

Added

• Transactions & Accounts — a real actuals ledger. Browse and manage your transactions, and manage your chart of accounts (with safe, guarded deletes). Manual entry gets full add/edit/delete; everything stays read-only and safe while a scenario is active.
• Stripe integration — connect Stripe with a read-only restricted key and Burnless pulls in your real transactions: revenue and payment-processing fees, split correctly, with vendor names attached. It backfills on connect, syncs hourly, and has a Sync now button for an immediate refresh. Find it under Connections → Integrations.
• Integration framework — a general connector spine behind Stripe, so more sources can plug in the same way: paste a key, data flows in.
• calculate AI tool — the AI companion can now do exact arithmetic on your numbers instead of estimating.
• Extensibility foundation — a registry-driven domain spine and pgvector-backed memory underpin the above (mostly internal groundwork for what's next).

Database

This release adds two tables (memory, integration_credentials) and a vector index. Migrations are additive and non-destructive and apply automatically on update — your existing data is untouched.

Install / update

curl -fsSL https://burnless.ai/install | sh   # new install
burnless update                                # existing install

Verify the download:

shasum -a 256 -c SHASUMS256.txt

Full changes: #11, #12, #13, #14.

v0.2.1

Quality-of-life fixes across the CLI, installer, AI, and self-host settings — bundles PRs #8 and #9.

Fixed

• burnless with no arguments now prints help instead of starting the app; the installer hands off via burnless start.
• Installer PATH — install.sh adds ~/.burnless/bin to your shell PATH automatically (shell-detected, idempotent), so burnless is available in a new shell.
• No more AI config prompt on start — burnless start no longer prompts to set up a provider. Use burnless provider add or the onboarding step.
• Your timezone, everywhere — onboarding captures your real timezone; the AI companion is now aware of the current date/time; and scheduled automations run on your company's timezone (DST-correct), instead of always UTC.
• Self-host credits display — the AI dashboard no longer shows a misleading "500 credits" cap when you bring your own AI (credits are a cloud concept; self-host AI was never actually throttled).
• Landing copy button — the install-command "copy" button works, with a graceful fallback when the clipboard is blocked.

Install / update

curl -fsSL https://burnless.ai/install | sh   # new install
burnless update                                # existing install

Full changes: #8, #9.

v0.2.0

AI chat reliability rebuild plus new scenario capabilities for the assistant.

Highlights

• Rebuilt AI chat persistence on an append-only event log. Conversations that span multiple tool steps or pause for confirmation now resume reliably. This fixes a bug where completed tool calls could be dropped on resume — causing the assistant to loop or repeat actions. Chat history, the live pending-action card, and chat exports now all read from this single source of truth.
• The assistant can work inside scenarios mid-conversation. It can create a scenario, make changes within it, and exit again — with every write correctly targeted to the active scenario, even when the scenario is activated partway through a turn.

Improvements

• Configurable AI limits (self-host). The tool-iteration cap and output-token limit are now environment-configurable, and output is uncapped by default so longer answers aren't truncated. See the new entries in .env.example.
• Tool-loop convergence guard. A two-tier guard stops the assistant from getting stuck in repetitive tool-call loops.

Fixes

• The assistant now shows a friendly message when a turn fails, instead of surfacing the raw model error.
• Finishing the onboarding wizard now triggers insights and data-freshness checks immediately, so the dashboard is current right after setup.

⚠️ Database migration (breaking)

This release drops the retired ai_messages and ai_pending_actions tables — chat now lives in the new append-only ai_turn_events log. burnless update applies the migration automatically for self-hosters. Conversation history previously stored in those tables is not carried into the new log; new conversations are unaffected.

Install / update

curl -fsSL https://burnless.ai/install | sh   # new install
burnless update                                # existing install

Note

For the most reliable AI experience, use a capable model for the assistant — very small models can struggle with the full tool set. Configure your provider in Settings → AI or during onboarding.

v0.1.3

Reliability and self-host quality fixes.

Fixes

• AI chat reliability. Resolved intermittent "couldn't generate a response" failures. All AI calls now go through a single resilient provider layer (automatic retry, usage tracking, and logging) regardless of how the provider is configured, and transient empty model responses are retried instead of failing the turn.
• Server log visibility. burnless start now prints operational logs — per-request status, AI request OK/FAIL with retry/circuit info, and errors — so self-hosters can see what's happening. (No secrets or message contents are logged.)
• Dashboard after onboarding. Finishing the setup wizard now opens the real dashboard immediately instead of briefly showing "Create Your First Scenario."
• AI automations. The assistant no longer claims an automation is created before you confirm it, and a newly confirmed automation now appears in the list right away.
• Localhost access (self-host). Visiting your instance at localhost no longer fails every action with a CSRF "invalid origin" error — localhost, 127.0.0.1, and ::1 are now treated as the same loopback origin.

Install / update

```sh
curl -fsSL https://burnless.ai/install | sh # new install
burnless update # existing install
```

Note

For the most reliable AI experience, use a capable model for the assistant (very small models can struggle with the full tool set). Configure your provider in Settings → AI or during onboarding.

v0.1.2

chore(release): v0.1.2

Ships the cross-platform installer + always-vendor Node + interactive CLI +
AI-onboarding fixes (PR #2). Notably fixes the Node-20/Alpine artifact crash
(v0.1.1 was broken on Alpine).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

v0.1.1

burnless v0.1.1 — first public release