tesseract-agent 0.11.25 → 0.11.27

package/README.md

@@ -36,7 +36,7 @@ The service auto-starts at login/boot and restarts on crash: **launchd** LaunchA
 on macOS, **systemd `--user`** unit on Linux (with `loginctl enable-linger` so it
 survives logout), **Task Scheduler** ONLOGON task on Windows. Run `install` as your
 normal user — not `sudo` (a per-user service needs your own session, and finds your
-credentials + `~/.claude`).
+credentials and your frameworks' own sign-ins in your home directory).
 
 ## Multiple agents under one user
 
@@ -68,5 +68,36 @@ docker run --rm \
 Or pre-provision credentials and mount them (persist `~/.tesseract` or set
 `CCB_DAEMON_DIR` to a mounted volume), then just `tesseract-agent run`.
 
-The agent runs under the machine's own Claude credentials — a `claude` login in
-`~/.claude`, or an `ANTHROPIC_API_KEY` in the environment.
+## Frameworks and models
+
+The agent isn't tied to one agent framework or one model vendor. You pick the
+**framework** that runs the agent, and the **provider** whose models it talks to.
+
+**Framework.** `run --engine <id>` picks it: `claude-code` (the default),
+`openai-codex`, `opencode`, `gemini-cli`, `qwen-code`, `kimi`, `cursor`, `hermes`,
+`openclaw` and the rest Tesseract supports. The framework's own CLI must be on the
+machine. `tesseract-agent assist install <framework>` installs it, and `--engine-path`
+points at a copy that isn't on `PATH`.
+
+**Provider.** A runner config pairs a framework with a model provider and its
+credential: Anthropic, OpenAI, Google, OpenRouter, DeepSeek, Moonshot, Z.ai,
+MiniMax, xAI, Azure OpenAI, Bedrock, Vertex, a local Ollama or LM Studio, or any
+OpenAI-, Anthropic- or Gemini-compatible endpoint (`--provider custom-openai
+--base-url …`).
+
+```bash
+tesseract-agent runner add --engine opencode --provider openrouter \
+  --secret '$OPENROUTER_API_KEY' --model <model> --select
+tesseract-agent runner login claude      # add a Claude subscription
+tesseract-agent runner list
+```
+
+A secret given as `$VAR` is read from the environment when the agent runs, so it is
+never written to disk. Add several configs and the agent rotates between them when
+one hits a rate limit, a quota or an auth error, across providers and frameworks.
+The running agent picks up changes on its next turn.
+
+With no runner config, the framework uses whatever sign-in it already has on this
+machine: a `claude` login or `ANTHROPIC_API_KEY` for Claude Code, a `codex` login
+or `OPENAI_API_KEY` for Codex, and so on. `--engine-env KEY=VAL` passes extra
+credentials to the framework.