recall-rag 0.14.0

Memory that abstains instead of guessing.
RE-call is agent memory on your own PostgreSQL with pgvector. Each result carries a verdict, confidence, and provenance, while unsupported questions are refused instead of answered by the nearest match.

What it is  ·  For agents  ·  How it works  ·  Try the demo  ·  Install and integrate

What is RE-call

RE-call is agent memory on your own PostgreSQL database. It indexes source documents with pgvector and keeps validity, lineage, confidence, and provenance attached to every result.

Vector search returns nearby text. RE-call also checks whether that text is current, supported, and trustworthy enough for the query. A superseded claim is marked superseded; a result that does not clear the calibrated trust gate becomes ABSTAIN with a reason. Declared supersession makes the current memory win over stale but similar memory.

The default path uses local embeddings plus hybrid dense and Postgres full text retrieval. It needs no memory-layer LLM call. Additional retrieval, reasoning, and structured-fact modules are optional.

For the agent reading this, RE-call for agents explains the experience loop and how to use persistent memory as a trustworthy past rather than an unverified scratchpad.

How it works

RE-call builds an immutable, calibrated generation once and serves queries against that pinned generation. The solid path is the default. Dashed arrows mark the optional reasoning graph path or generation bindings. Nothing bypasses the generation or trust boundary.

flowchart TB
    subgraph BUILD["1. Build and certify a generation"]
        direction LR
        SOURCE["Corpus<br/>memo files + frontmatter"] --> PREP["Validate + manifest<br/>chunk + embed"]
        PREP --> GEN[("Immutable generation<br/>PostgreSQL + pgvector")]
        GEN --> GRAPH["Evidence Graph V1<br/>generation bound"]
        GEN --> CAL["Calibration<br/>published for this generation"]
    end

    subgraph READ["2. Serve every query"]
        direction LR
        CLIENT["Agent or application"] --> ACCESS["Python API · CLI · MCP"]
        ACCESS --> PRECHECK["Pin active generation<br/>require published calibration"]
        PRECHECK -->|"ready"| RETRIEVE["Hybrid retrieval<br/>dense vectors + Postgres full text"]
        PRECHECK -->|"not ready"| REFUSE["REFUSE before retrieval<br/>reason returned"]
        RETRIEVE --> DIRECT["Direct candidates"]
        DIRECT --> GATE{"Calibrated trust gate<br/>validity + supersession + confidence"}
        ACCESS -. "explicit reasoning query" .-> OPTION["graph_expansion<br/>auto | off | one-hop"]
        OPTION -. "auto: bounded one hop<br/>for every nonempty query" .-> EXPAND["Read graph neighbors<br/>same generation"]
        DIRECT -. "when option resolves to one hop" .-> EXPAND
        GRAPH -. "generation binding" .-> EXPAND
        EXPAND --> MERGE["Merge candidates<br/>direct hits stay first"]
        MERGE --> GATE
        GATE -->|"trusted"| TRUSTED["Trusted evidence<br/>verdict + provenance"]
        GATE -->|"unsupported or stale"| ABSTAIN["ABSTAIN<br/>reason returned"]
        TRUSTED --> ANSWER["Optional cited answer<br/>review or abstain"]
        TRUSTED --> CARDS["Optional evidence cards<br/>source digest + lineage"]
        CARDS --> CONTROLLER["Provenance controller<br/>review + recheck"]
        CONTROLLER --> LEDGER[("Append only fact ledger<br/>asserted · refused · superseded")]
    end

    GEN -. "active snapshot" .-> PRECHECK
    CAL -. "threshold bound to generation" .-> PRECHECK

    classDef core fill:#e8f3ff,stroke:#2b6cb0,color:#102a43,stroke-width:1px;
    classDef trust fill:#e8f5e9,stroke:#2f855a,color:#163b27,stroke-width:1px;
    classDef optional fill:#fff8e1,stroke:#b7791f,color:#5f370e,stroke-width:1px;
    classDef stop fill:#fff1f2,stroke:#c53030,color:#63171b,stroke-width:1px;
    class SOURCE,PREP,GEN,GRAPH,CAL,CLIENT,ACCESS,PRECHECK,RETRIEVE,DIRECT,OPTION,EXPAND,MERGE core;
    class GATE,TRUSTED trust;
    class ANSWER,CARDS,CONTROLLER,LEDGER optional;
    class REFUSE,ABSTAIN stop;

Ordinary recall search follows the direct path. Explicit reasoning accepts graph_expansion: auto is the default and resolves to bounded one hop expansion for every nonempty query; off keeps direct retrieval only; one-hop forces the graph path. The CLI uses --graph-expansion auto|off|one-hop; the MCP tool uses graph_expansion="auto"|"off"|"one_hop". Graph neighbors are generation bound, direct candidates remain first, and expanded candidates must clear the same trust boundary before they can support a cited answer.

The opt in choices attach to different points in the system:

For details, see the architecture writeup, provenance controller, and API reference.

Quickstart

Prerequisites: Python 3.11 or newer, Docker, and a Docker installation able to run PostgreSQL with pgvector.

pip install "recall-rag[fastembed]"
recall quickstart

The demo starts a throwaway database, indexes a small corpus included in the package, and runs three searches. It includes a normal answer, a stale claim that is returned as superseded, and a question that is refused. The demo uses development trust and changes no personal files.

Remove the demo database when finished:

recall quickstart --remove

Already have PostgreSQL with pgvector? Use recall quickstart --existing-dsn <dsn> instead. The demo is intentionally separate from a real install and is not calibrated for your data.

Install and integrate

For your own corpus, provide PostgreSQL with pgvector and run the guided setup wizard after installing recall-rag[fastembed]:

recall setup

It applies the schema, asks for the embedder and retrieval options, indexes the corpus, offers calibration, and registers the selected agent integration. When the wizard asks whether to calibrate, use a labeled query file that refers to the corpus you are installing. Calibration fitted to the bundled demo is only an example, not a certification for your data. The schema uses an ordered SQL migration path and pre-tenancy tables are migrated in place.

For Docker, an existing database, headless provisioning, manual calibration, and troubleshooting, see docs/INSTALLATION.md and docs/WIZARD.md.

Choose an integration

Claude Code

Inside Claude Code, install the plugin after installing the Python package:

/plugin marketplace add GiulioDER/RE-call
/plugin install recall@re-call

The plugin supplies the MCP server, memory search skill, and lifecycle hooks. recall setup still needs to run against the project and database that Claude should use. The plugin keeps credentials out of the repository. Details and manual wiring are in plugin/README.md.

Codex

Run recall setup from the project. When Codex is detected, setup installs the Codex MCP server, plugin bundle, memory skills, and hooks into the user configuration. Restart Codex afterward. The Codex and Claude Code integrations share the same memo format and trust layer. See docs/CODEX_RECALL_INTEGRATION.md.

Claude Agent SDK

The SDK integration runs the same tools in process and does not start an MCP server:

from recall_agent import RecallAgentMemory

with RecallAgentMemory.from_env() as memory:
    options = memory.options()

See docs/USING_WITH_AGENT_SDK.md for the complete example and write-tool boundaries.

LangChain and LlamaIndex

Both adapters use the same trusted retrieval path. If trust abstains, they return no document by default, and returned documents retain verdict, confidence, cosine, and supersession metadata. See docs/API.md for the supported classes and methods.

If an install is not working

recall doctor

The doctor checks the interpreter, console scripts, embedder, Docker, database, pgvector, schema, configured table and tenant, calibration, and agent registration. It changes nothing and prints the repair command for each problem.

Product surface

RE-call is not a hosted memory service, dashboard, or automatic truth extractor. It does not rewrite corpus metadata from an agent's inference. Reasoning is opt in, citation constrained, and review aware. See docs/PRODUCTION.md for deployment boundaries.

Read next

Published numbers are tied to committed artifacts, and the claim gate checks them in CI. Benchmark interpretation and limits belong in docs/EVIDENCE.md, not in this overview.

Citation

If you describe RE-call in a paper, post, talk, or README of your own, cite the project and credit Giulio D'Erme. Use CITATION.cff as the canonical citation source.

License

RE-call is source available under the PolyForm Noncommercial License 1.0.0. Personal, educational, and noncommercial research use is permitted. Commercial use requires a separate written license from the copyright holder. See COMMERCIAL_LICENSE.md for the boundary between permitted use and commercial licensing, and preserve NOTICE when redistributing the software.