v0.20.3

Full Changelog: v0.20.2...v0.20.3

Cercano v0.20.2 — First Standalone Agent Beta

First beta of the standalone Cercano agent

Cercano v0.20.2 is the first beta of Cercano as a standalone agent—the next step beyond its earlier coprocessor role. It now ships as an agent plus a terminal client, so you can run Cercano directly without a source checkout or development launcher.

Each archive includes two executables:

• cercano — the agent that performs the work.
• cercano-cli — the terminal client. Start here; it launches the agent when needed.

This beta is available for macOS Apple Silicon and Linux x86-64. We welcome testing and bug reports, especially around installation, first-run setup, and everyday agent use. Expect rough edges: Linux is experimental, Windows is not included in this release, and full cross-platform runtime validation is still in progress. Models and provider credentials are configured separately and are not bundled.

Downloads

• Linux x86-64: unsigned experimental archive containing cercano and cercano-cli. Native Ubuntu build, isolated version smoke tests, and process-helper tests passed. Full runtime and distribution support is still being validated; this is not a Debian/APT package.
• macOS Apple Silicon: Developer ID signed and Apple-notarized archive containing both binaries. Build, signing, notarization and archive verification passed.
• SHA-256 checksum files are provided for both archives.

Windows is not included: its workflow-test harness failed before compilation. The overall rehearsal run is red for that reason. The successful Linux/macOS artifacts are published separately with explicit operator authorization; no Windows tests were bypassed or claimed to pass.

Artifacts are the unchanged outputs of CI run 36798835095, commit 15ae973ae49105c14b7c229a3e44c2f6c1297021.

Linux quick start

sha256sum -c cercano-0.20.2-linux-x64.tar.gz.sha256
tar -xzf cercano-0.20.2-linux-x64.tar.gz
./cercano-0.20.2-linux-x64/bin/cercano-cli

Models and provider credentials are not included. Headless credential storage and full agent/model operation remain under validation.

v0.20.0

What's Changed

• fix(llamaserver): kill the whole process group when install is cancelled by @keithballinger in #24
• fix(anthropicauth): escape reflected error in loopback callback page by @keithballinger in #21
• fix(localruntime): claim download slot atomically in DownloadModel by @keithballinger in #23
• fix(worker): reuse one token source per profile in resolveCredential by @keithballinger in #22
• Updates to allow building on Windows hosts by @ctacke in #26
• More windows-based changes by @ctacke in #27
• feat(release): integrate signed macOS standalone release pipeline by @bryancostanich in #33

New Contributors

• @ctacke made their first contribution in #26
• @bryancostanich made their first contribution in #33

Full Changelog: https://github.com/bryancostanich/Cercano/commits/v0.20.0

v0.10.0

Full Changelog: v0.9.3...v0.10.0

v0.9.3

Full Changelog: v0.9.2...v0.9.3

v0.9.2

Full Changelog: v0.9.1...v0.9.2

v0.9.1

Full Changelog: v0.9.0...v0.9.1

v0.9.0

Full Changelog: v0.8.1...v0.9.0

v0.8.1

Cercano v0.8.1

Fix: Savings Estimation Accuracy

Local model output tokens were incorrectly counted as cloud overhead, making savings appear near zero. Fixed — overhead is now just ~100 tokens per call (tool call formatting). A deep research run that keeps 56K tokens out of the cloud now correctly reports ~55K net savings instead of 571.

Upgrade

brew upgrade cercano

Full Changelog: v0.8.0...v0.8.1

v0.8.0

Cercano v0.8.0

New: cercano_deep_research — Multi-Source Research Tool

Give it a topic and intent, and it systematically researches across 25+ sources:

• Source planning — local model picks relevant sources (PubMed, arXiv, GitHub, Wired, MIT Tech Review, etc.) based on your topic
• Academic APIs — PubMed and arXiv searched via their free public APIs
• Site-scoped search — DuckDuckGo with site targeting for everything else
• Rich analysis — each finding gets a detailed summary with key facts, relevance scoring (1-5 stars), impact rating, and actionable "how to use" notes
• Reference chasing — discovers cited works from findings and analyzes them too (1-hop, up to 50)
• Synthesis — executive summary, narrative, contradiction detection, gap analysis, recommended reading order, follow-up suggestions
• Multi-file output — writes to scratch/research/<topic>/ in your project with README, individual finding files, synthesis, and source plan
• Crash recovery — checkpoints intermediate state, resumes on retry

The entire pipeline runs locally. The host agent gets only a 4-line summary.

Upgrade

brew upgrade cercano

Full Changelog: v0.7.2...v0.8.0