<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: synapse builder</title>
    <description>The latest articles on DEV Community by synapse builder (@synapsebuiler).</description>
    <link>https://dev.to/synapsebuiler</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3994465%2F8a0322dc-bd0c-4ec2-9b72-4bbddb5698cc.png</url>
      <title>DEV Community: synapse builder</title>
      <link>https://dev.to/synapsebuiler</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/synapsebuiler"/>
    <language>en</language>
    <item>
      <title>Tera Pilot — a self-hosted, vendor-neutral coding agent you can actually verify (looking for testers 🙏)</title>
      <dc:creator>synapse builder</dc:creator>
      <pubDate>Wed, 26 Aug 2026 09:44:27 +0000</pubDate>
      <link>https://dev.to/synapsebuiler/tera-pilot-a-self-hosted-vendor-neutral-coding-agent-you-can-actually-verify-looking-for-1och</link>
      <guid>https://dev.to/synapsebuiler/tera-pilot-a-self-hosted-vendor-neutral-coding-agent-you-can-actually-verify-looking-for-1och</guid>
      <description>&lt;p&gt;If you've been trying out AI coding agents and kept running into the same three questions — &lt;em&gt;"where is my code actually going?"&lt;/em&gt;, &lt;em&gt;"can I use a local model instead of paying per token?"&lt;/em&gt;, and &lt;em&gt;"how do I know what it did to my repo?"&lt;/em&gt; — I want to show you &lt;strong&gt;&lt;a href="https://github.com/ilyaosovskoi/tera-pilot" rel="noopener noreferrer"&gt;Tera Pilot&lt;/a&gt;&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;It's an MIT-licensed, self-hosted coding agent built around one idea: &lt;strong&gt;trust through evidence, not marketing claims.&lt;/strong&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What it actually is
&lt;/h2&gt;

&lt;p&gt;Tera Pilot is a Textual-based TUI-first coding agent (with a web UI, a REST/SSE daemon, and MCP/ACP support too) that runs a classic Plan → Explore → Act → Verify → Report loop against your repo. What makes it different from the usual "wrapper around an LLM API":&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Bring your own key or go fully offline.&lt;/strong&gt; 16+ providers (Anthropic, OpenAI, Gemini, DeepSeek, Groq, xAI, Mistral, Cerebras, Together, Fireworks, SambaNova, Nvidia NIM, OpenRouter…) or point it at Ollama / LM Studio and never send code to the cloud.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Sandboxed by default.&lt;/strong&gt; Workspace-scoped file access, command allow/deny policy, an OS-level sandbox (macOS &lt;code&gt;sandbox-exec&lt;/code&gt;, Linux &lt;code&gt;bubblewrap&lt;/code&gt;) that denies network and blocks reads of &lt;code&gt;~/.ssh&lt;/code&gt;, &lt;code&gt;~/.aws&lt;/code&gt;, etc. during command execution.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A "Guardian" that reviews risky tool calls&lt;/strong&gt; and can approve/reject/modify them — and fails &lt;em&gt;closed&lt;/em&gt; on errors instead of silently letting things through.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Signed, tamper-evident audit trail.&lt;/strong&gt; Every tool call is logged; you can export it with Ed25519 signatures + a SHA-256 hash chain and verify it on a completely different machine.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;It publishes its own security testing.&lt;/strong&gt; 671 tests total, 210 of them security/sandbox/policy tests, plus a documented list of five real vulnerabilities they found and fixed via offensive testing (git alias/config shell escapes, an SSRF-adjacent CORS bug, npm script execution, etc.) — with an explicit "what we're NOT claiming" section instead of a hand-wavy security badge.
There's also a reproducible eval harness (58 tasks: bug fixes, refactors, test repair, code review, adversarial security tasks) with results checked in, including a run against a fully local 2.6B model in LM Studio.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Why I'm posting this
&lt;/h2&gt;

&lt;p&gt;The project is honest about being in a &lt;strong&gt;testing phase&lt;/strong&gt; — the README says so upfront, no marketing gloss. The maintainer is looking for people to actually run it against real repos, hit edge cases, and file issues. That's a refreshing amount of candor for a project asking for stars, so I wanted to help it find the right audience: people who care about local-first tooling, sandboxing, and not vibing on trust claims.&lt;/p&gt;

&lt;h2&gt;
  
  
  Quick start
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npm &lt;span class="nb"&gt;install&lt;/span&gt; &lt;span class="nt"&gt;-g&lt;/span&gt; tera-pilot
tera-pilot-tui        &lt;span class="c"&gt;# full-screen terminal UI, the primary way to use it&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;If npm gives you trouble (it's noted as the rough edge right now), the source install is one extra step and more reliable:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;git clone https://github.com/ilyaosovskoi/tera-pilot.git
&lt;span class="nb"&gt;cd &lt;/span&gt;tera-pilot
python3 &lt;span class="nt"&gt;-m&lt;/span&gt; venv .venv &lt;span class="o"&gt;&amp;amp;&amp;amp;&lt;/span&gt; &lt;span class="nb"&gt;source&lt;/span&gt; .venv/bin/activate
pip &lt;span class="nb"&gt;install&lt;/span&gt; &lt;span class="nt"&gt;-e&lt;/span&gt; &lt;span class="nb"&gt;.&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Then run &lt;code&gt;tera-pilot doctor&lt;/code&gt; — it checks your Python version, provider keys, local model servers, and workspace in one shot, so you know immediately if something's misconfigured.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who this is for
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Devs who want a coding agent but don't want their private repo leaving their machine&lt;/li&gt;
&lt;li&gt;Anyone who wants to try local models (Ollama/LM Studio) as a serious agent backend, not a toy&lt;/li&gt;
&lt;li&gt;People who actually read threat models — there's a public one, plus a security report with real numbers, not just a shield emoji in the README&lt;/li&gt;
&lt;li&gt;
&lt;p&gt;Teams evaluating self-hosted alternatives to hosted agent products before committing&lt;/p&gt;
&lt;h2&gt;
  
  
  How you can help
&lt;/h2&gt;
&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;Clone it, run &lt;code&gt;tera-pilot doctor&lt;/code&gt;, and try a real task on a real repo.&lt;/strong&gt; Bug reports and rough-edge reports are exactly what the project needs right now.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;&lt;strong&gt;If it does what it says on the tin, a star helps a lot&lt;/strong&gt; — it's a solo/small-team effort trying to get real-world signal before a public release.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Check the &lt;a href="https://github.com/ilyaosovskoi/tera-pilot/blob/main/SECURITY_TEST_REPORT.md" rel="noopener noreferrer"&gt;security report&lt;/a&gt; and &lt;a href="https://github.com/ilyaosovskoi/tera-pilot/blob/main/THREAT_MODEL.md" rel="noopener noreferrer"&gt;threat model&lt;/a&gt; if that's your kind of reading material — feedback from people who actually think about sandboxing is valuable here.&lt;br&gt;
🔗 Repo: &lt;a href="https://github.com/ilyaosovskoi/tera-pilot" rel="noopener noreferrer"&gt;https://github.com/ilyaosovskoi/tera-pilot&lt;/a&gt;&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Would love to hear what breaks for you.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>opensource</category>
      <category>python</category>
      <category>devtools</category>
    </item>
  </channel>
</rss>
